KR102434112B1 - Method and apparatus for generating disease prediction ai model, and system and method for predicting user-customized disease using the same - Google Patents

Abstract

An embodiment of the present invention provides an apparatus for generating an artificial intelligence model that provides customized disease prediction information to a user. The apparatus includes a memory for storing a disease prediction model generation program and a processor for executing the program stored in the memory. The processor is configured to: by executing the disease prediction model generation program, pre-process learning data including medical information of each of a plurality of insurers to generate a learning data set; based on test data including some of the training data included in the training data set, perform performance evaluation including accuracy evaluation for each of the plurality of artificial intelligence to select an artificial intelligence model with the highest performance index among the plurality of artificial intelligence models; set personal data, medical treatment data, and lifestyle data of insurers included in the learning data set as input variables; and create a user-customized disease prediction model by setting the disease occurrence data of the insurers included in the learning data set as an output variable and learning the selected artificial intelligence model based on the learning data set.

Description

Method and apparatus for generating a disease prediction model, and a user-customized disease prediction system and method using the same

The present invention relates to a method and apparatus for generating a disease prediction model, and a user-customized disease prediction system and method using the same. A method and apparatus for generating an artificial intelligence model, and a user-customized disease prediction system and method using the same.

Recently, in modern society, the incidence of diseases such as cancer, heart disease, cerebrovascular disease, liver disease, diabetes, high blood pressure, thyroid gland, renal disease, glaucoma, and arteriosclerosis continues to increase, and the age of onset is also decreasing.

Conventionally, in order to determine whether the general public has the above diseases, it is necessary to visit a medical institution and receive periodic health check-ups. If the general public does not regularly visit medical institutions and do not receive health check-ups, they may not be able to detect the presence of the disease and may miss the time to treat the disease, which may exacerbate the disease. Aggravation of the disease can lead to a crisis of life itself and negative emotions beyond the physical problems caused by the disease itself.

In order to prevent the above problems, it is necessary for the general public to identify diseases that are most likely to contract in the future based on their lifestyle and medical history information, and a technology that allows them to easily receive preventive measures to prevent the disease from anywhere. to be.

Republic of Korea Patent Publication No. 10-2188766 (2020.12.1.) (Title of the invention: AI-based healthcare service providing device)

An object of the present invention is to provide an apparatus and method for generating a disease prediction model that provides user-customized disease prediction information.

In addition, the present invention inputs the user's history data and lifestyle data into the disease prediction model, analyzes the risk of disease predicted according to risk factors among the above data, and provides symptoms, self-diagnosis and prevention methods for the predicted disease to the user. Another technical task is to provide a user-customized disease prediction system and method for

Another technical object of the present invention is to provide a user-customized disease prediction system and method for recommending to a user an insurance product that guarantees a disease predicted through a disease prediction model.

The technical problems to be achieved by the present invention are not limited to the above technical problems, and other technical problems of the present invention may be derived from the following description.

As a technical means for solving the above technical problem, according to the first aspect of the present invention, there is provided a method of generating an artificial intelligence model that provides customized disease prediction information to a user. In the method, the apparatus comprises: A training data set is generated by preprocessing training data including medical information of each of a plurality of insurers, and each of a plurality of artificial intelligence models is based on test data including some of the training data included in the training data set. Selects an AI model having the highest performance index among the plurality of AI models by performing performance evaluation including accuracy evaluation for Habit data is set as an input variable, disease occurrence data of the insurers included in the learning data set is set as an output variable, and the selected artificial intelligence model is trained based on the learning data set to create a user-customized disease prediction model. comprising the steps of creating

In addition, as a technical means for solving the above-described technical problem, according to a second aspect of the present invention, an apparatus for generating a user-customized disease prediction model is provided. The apparatus includes a memory storing a disease prediction model generation program and a processor executing the program stored in the memory, wherein the processor executes the disease prediction model generation program to include medical information of each of a plurality of insurers pre-processing the training data to generate a training data set, and perform performance evaluation including accuracy evaluation for each of a plurality of artificial intelligence models based on test data including some of the training data included in the training data set to select an artificial intelligence model having the highest performance index among the plurality of artificial intelligence models, and set personal data, medical treatment data, and lifestyle data of the insurers included in the learning data set as input variables, and setting the disease occurrence data of the insurers included in the learning data set as an output variable, and learning the selected artificial intelligence model based on the learning data set to generate a user-customized disease prediction model.

In addition, as a technical means for solving the above technical problem, according to a third aspect of the present invention, there is provided a method of providing customized disease prediction information to a user through a communication connection between a terminal and a server. In this method, the server receives user input information from the terminal, and based on a learning data set including personal data, medical treatment data, lifestyle data, and disease occurrence data of insurers received from the National Health Insurance Corporation server. , generating user disease prediction information based on the user input information by using the disease prediction model learned by using the personal data, medical treatment data, and lifestyle data as input variables and using the disease occurrence data as output variables, and and providing, by the server, the user disease prediction information to the terminal.

In addition, as a technical means for solving the above technical problem, according to a fourth aspect of the present invention, a system for providing customized disease prediction information to a user through a communication connection with a terminal is provided. The system includes a communication module that transmits and receives information to and from the terminal, a memory that stores a program for providing user-customized disease prediction information, and a processor that executes the program stored in the memory. The processor executes the user-customized disease prediction information providing program, receives user input information from the terminal, and includes personal data, medical treatment data, lifestyle data, and disease occurrence data of the insurers received from the National Health Insurance Corporation server. Predicting user disease based on the user input information by using the disease prediction model learned by using the personal data, medical treatment data, and lifestyle data as input variables and the disease occurrence data as output variables based on the learning data set and generating information, and providing the user disease prediction information to the terminal.

According to the above-described problem solving means of the present invention, the present invention can provide an apparatus for generating a disease prediction model that provides user-customized disease prediction information.

In addition, the present invention inputs the user's history data and lifestyle data into the disease prediction model, analyzes the risk of disease predicted according to risk factors among the above data, and provides symptoms, self-diagnosis and prevention methods for the predicted disease to the user. We can provide a system that provides

In addition, the present invention may provide a system for recommending an insurance product that guarantees a disease predicted through a disease prediction model to a user.

1 is a block diagram illustrating a configuration of an apparatus for generating a disease prediction model according to an embodiment of the present invention.
2 is a flowchart illustrating a sequence of a method for generating a disease prediction model according to another embodiment of the present invention.
3 to 5 are diagrams illustrating detailed processes for some steps of the method for generating a disease prediction model shown in FIG. 2 .
6 to 8 are diagrams illustrating the results of analyzing the correlation between the learning data integrated in the learning data integration and pre-processing steps of FIG. 3 .
9 is a diagram illustrating a user-customized disease prediction system according to another embodiment of the present invention.
FIG. 10 is a block diagram illustrating the configuration of the user-customized disease prediction system shown in FIG. 9 .
11 is a block diagram showing the configuration of the terminal shown in FIG.
12 is a flowchart illustrating a user-customized disease prediction method according to another embodiment of the present invention.
FIG. 13 is a diagram illustrating an additional step of the method for predicting a user-customized disease shown in FIG. 12 .

Hereinafter, the present invention will be described in detail with reference to the accompanying drawings. However, the present invention may be implemented in various different forms, and is not limited to the embodiments described herein. In addition, the accompanying drawings are only for easy understanding of the embodiments disclosed in the present specification, and the technical ideas disclosed in the present specification are not limited by the accompanying drawings. All terms including technical terms and scientific terms used herein should be interpreted as meanings commonly understood by those of ordinary skill in the art to which the present invention belongs. Terms defined in the dictionary should be interpreted as having additional meanings consistent with the related technical literature and the presently disclosed content, and are not interpreted in a very ideal or limiting sense unless otherwise defined.

In order to clearly explain the present invention in the drawings, parts irrelevant to the description are omitted, and the size, shape, and shape of each component shown in the drawings may be variously modified. The same/similar reference numerals are attached to the same/similar parts throughout the specification.

The suffixes “module” and “part” for components used in the following description are given or mixed in consideration of only the ease of writing the specification, and do not have distinct meanings or roles by themselves. In addition, in describing the embodiments disclosed in the present specification, when it is determined that detailed descriptions of related known technologies may obscure the gist of the embodiments disclosed in the present specification, detailed descriptions thereof are omitted.

Throughout the specification, when a part is said to be "connected (connected, contacted, or coupled)" with another part, this means not only when it is "directly connected (connected, contacted, or coupled)" but also refers to another member in the middle. It also includes the case of "indirectly connected (connected, contacted, or combined)" between them. Also, when a part "includes (provides or provides)" a component, it does not exclude other components unless otherwise stated, but further "includes (provides or provides)" other components. means you can

As used herein, terms indicating ordinal numbers such as first, second, etc. are used only for the purpose of distinguishing one element from other elements, and do not limit the order or relationship of the elements. For example, a first component of the present invention may be referred to as a second component, and similarly, a second component may also be referred to as a first component. Forms of the singular expression used herein should be construed to include forms of the plural expression as well, unless the meaning is clearly indicated to the contrary.

1 is a block diagram illustrating a configuration of an apparatus 100 for generating a disease prediction model according to an embodiment of the present invention.

Referring to FIG. 1 , the apparatus 100 for generating a disease prediction model may include a memory 110 and a processor 130 , and may further include a communication module 120 .

The memory 110 stores a disease prediction model generation program. Also, the memory 110 may be configured to store at least one of data generated according to the execution of the processor 130 . The memory 110 should be interpreted as a generic term for a non-volatile storage device that continuously maintains stored information even when power is not supplied, and a volatile storage device that requires power to maintain the stored information. The memory 110 may perform a function of temporarily or permanently storing data processed by the processor 130 . The memory 110 may include magnetic storage media or flash storage media in addition to the volatile storage device that requires power to maintain stored information, but the scope of the present invention is not limited thereto. not.

The communication module 120 may transmit/receive data necessary for generating a disease prediction model by performing information transmission/reception with an external device or server. The communication module 120 may include a device including hardware and software necessary for transmitting and receiving signals such as control signals or data signals through wired/wireless connection with other network devices.

The processor 130 is configured to execute a disease prediction model generation program stored in the memory 110 . The processor 130 may include various types of devices for controlling and processing data. The processor 130 may refer to a data processing device embedded in hardware having a physically structured circuit to perform a function expressed as a code or an instruction included in a program. In one example, the processor 130 includes a microprocessor, a central processing unit (CPU), a processor core, a multiprocessor, an application-specific integrated circuit (ASIC), an FPGA ( field programmable gate array), but the scope of the present invention is not limited thereto.

The processor 130 is configured to execute the disease prediction model generation program to perform the following functions and procedures.

The processor 130 pre-processes the training data including medical information of each of the plurality of insurers to generate a training data set. The processor 130 receives learning data including medical information from the National Health Insurance Corporation (KNHIS). The medical information may include insurance eligibility data, insurance premium data, birth data, death data, medical treatment data, disease history data, and health checkup data of each of the plurality of insurers. For example, the medical information may be in the form of an Excel file classified into several. The training data is unstructured data, and the training data set obtained by preprocessing the training data is structured data.

When the processor 130 executes the disease prediction model generation program to pre-process the learning data, the treatment data, personal injury data, injury data, personal data, lifestyle data, and treatment data in the medical information are set as feature data that is the labeling target of the learning data set. more can be done The processor 130 may integrate the learning data into one, and extract each of the plurality of insurers' personal injury data, injury data, personal data, lifestyle data, and medical treatment data from the combined learning data. The processor 130 integrates main-injury data, wounded-injury data, personal data, lifestyle data, and medical treatment data through a common key. The processor 130 may analyze the correlation between the integrated data to determine data importance and data reliability. The main-injury data includes information on the health and illness that insurers have the greatest demand for treatment or testing. Injury data includes information on injuries and diseases that have affected patient care as a condition that occurred with or occurred during the insurers' treatment period. Personal data includes the gender and age of the insurer. The lifestyle data includes information on the amount of smoking, the number of days smoked, the number of days drunk, and the amount of alcohol consumed. Medical data includes at least one or more of the insurer's medical treatment period and treatment date, past disease history, recent disease history, family history, type of examination, medical history, visiting department, hepatitis B antigen holding and current drug use do. The common key may be the insurers' personal unique number (RN_INDI).

The processor 130 may analyze the correlation between the integrated data by using at least one of a Cramer V test, a random forest, and a linear regression. Cramer V test and Random Forest can identify the importance of data in the integrated data. Linear Regression can determine the reliability of data from the integrated data.

In more detail, the processor 130 determines that data having a Cramer V test value of 0.5 or more is main data having high importance, and data having a Cramer V test value of 0.3 or more and less than 0.5 is characteristic data having a low importance compared to the main data. can be set to For example, data having a Cramer V test value of 0.5 or more may be data for main and wounded soldiers. For example, data having a Cramer V test value of 0.3 or more and less than 0.5 may be past disease history and drug treatment history data among medical data.

The processor 130 may set data having a random forest value (IncNodePurity) of more than 100,000 as data having very high importance, and data having a random forest value (IncNodePurity) of more than 10,000 and less than or equal to 100,000 may be set as data having normal importance. For example, data having a Random Forest value (IncNodePurity) greater than 100,000 may be head injury data, injury disease data, age data of personal data, and visiting department data of treatment data. For example, data having a random forest value (IncNodePurity) greater than 10,000 and less than or equal to 100,000 may be lifestyle data, gender data of personal data, and medical treatment data.

When the linear regression value (P-value) is 0 or more and less than 0.5, the processor 130 may set the data to have high reliability. For example, data having a linear regression value (P-value) greater than or equal to 0 and less than 0.5 may be medical data and personal data excluding main-injury data, injury data, and family history data.

The processor 130 may generate learning data for generating an artificial intelligence model by referring to the data importance and reliability derived through the correlation analysis between the integrated data as described above, or set input variables and output variables of the artificial intelligence model. have. For example, the processor 130 may set a specific value (0 to 1) for each data through correlation analysis. In this case, the processor 130 may set the data to which a numerical value of 0.5 or more is assigned as an input variable and an output variable, or may set it as characteristic data that must be included in the learning data set. The processor 130 may generate a training data set by identifying an error in the learning data that is correlated and integrated, and correcting the error. The error correction of the training data may include at least one of missing value processing, outlier processing, data augmentation, and data unit conversion. Missing value processing can be corrected by removing the missing value from the training data or inserting a mode value. Data augmentation can be corrected by identifying Imbalanced Data of training data and applying Synthetic Minority Oversampling Technique (SMOTE). Data unit conversion is to fit data units by converting all data within a specific range through MinMax Scaling. For example, data unit conversion may be performed after data augmentation through SMOTE.

The processor 130 performs a performance evaluation on each of the plurality of AI models based on test data including some of the training data included in the training data set to obtain the highest performance index among the plurality of AI models. Select an artificial intelligence model with The test data includes test input data and test result data such as personal data, medical treatment data, and lifestyle data of any one of the plurality of insurers. The plurality of artificial intelligence models may include two or more of Random Forest, Decision Tree, XGBoost, Naive Bayes, and DNN. Performance evaluation includes accuracy evaluation, precision evaluation, recall evaluation, and F1 score evaluation for each of a plurality of artificial intelligence models. The performance index includes an accuracy index, a precision index, a recall index, and an F1 score index corresponding to the performance evaluation. For example, when the processor 130 performs performance evaluation for each of the plurality of artificial intelligence models, the performance index may appear as follows.

“Random Forest: 97% accuracy, 94% precision, 95% recall, 94% F1 value

Decision Tree: accuracy: 97%, recall: 93%, recall: 96%, F1 value: 95%

Naive Bayes: accuracy: 28%, recall: 32%, recall: 40%, F1 value: 27%

XGBoost: accuracy: 97%, recall: 93%, recall: 97%, F1 value: 95%

DNN (Deep Neural Network) : accuracy: 40%~94%”

The processor 130 selects the highest artificial intelligence model among the numerical values of the indicators including the F1 score in which the accuracy indicator, the precision indicator, the recall indicator, and the reliability are equally calculated among the plurality of artificial intelligence models. A criterion for determining whether the numerical value is high or low may be an average value calculated based on the numerical value of each of the indicators. The processor 130 may use a decision tree when selecting an AI model having the highest F1 score among a plurality of AI models.

The processor 130 may set personal data, medical treatment data, and lifestyle data of insurers included in the learning data set as input variables, and set disease occurrence data of insurers included in the learning data set as output variables. The input variable and the output variable are selected by referring to the data importance and reliability derived by the processor 130 through the correlation analysis between the integrated learning data. The input variables selected by referring to the correlation analysis between the integrated learning data include personal data, medical treatment data, and lifestyle data of insurers. The personal data may include at least one of age, gender, and date of birth. Personal data must include age and gender data. Medical data includes at least one or more of the insurer's medical treatment period and date, past disease history, recent disease history, type of examination, visiting department, family history, drug treatment history, hepatitis B antigen holding and current drug use can do. Medical data must include historical disease history, recent disease history, visiting department and drug treatment history data. The lifestyle data may include at least one of smoking status, smoking period, smoking amount, drinking status, drinking period, and drinking amount. Lifestyle data must include data on whether or not smoking, smoking period, smoking amount, drinking status, drinking period and drinking amount data. The disease occurrence data includes cancer occurrence data, heart disease occurrence data, brain disease occurrence data, liver disease occurrence data, diabetes occurrence data, hypertension occurrence data, thyroid disease occurrence data, renal gland occurrence data, glaucoma occurrence data, and arteriosclerosis occurrence data. may include. The output variable may be set by the processor 130 to include at least one or more data among data included in the disease occurrence data.

The processor 130 generates a user-customized disease prediction model by learning the AI model selected according to the AI model performance evaluation based on the training data set including the input variable and the output variable. The correlation between the main and injured disease data of the training data set and personal data, lifestyle data, and medical treatment data has the highest correlation with the output variable of the user-customized disease prediction model. For example, insurer A's main disease is lung cancer, and the injured disease is bronchial cancer. At this time, insurer A suffered from chronic cough disease within 1 year and smoked 5 days a week. When an input variable having the above treatment data and lifestyle data is input to a user-customized disease prediction model, the user-customized disease prediction model outputs lung cancer and bronchial cancer as output variables (disease prediction information), and outputs the output variables to cancer It can be categorized by occurrence data.

2 is a flowchart illustrating a sequence of a method for generating a disease prediction model according to another embodiment of the present invention, and FIGS. 3 to 5 are diagrams illustrating detailed processes for some steps of the method for generating a disease prediction model. Hereinafter, a method for generating a disease prediction model will be described with reference to FIGS. 2 to 5 . Each step of the method for generating a disease prediction model to be described below may be performed by the apparatus 100 for generating a disease prediction model previously described with reference to FIG. 1 . Accordingly, the contents of the embodiment of the present invention described above with reference to FIG. 1 can be equally applied to the embodiments of FIGS. 2 to 5 to be described below, and the contents overlapping with the description described in FIG. 1 will be omitted. do. The steps described in FIGS. 2 to 5 are not necessarily performed in order, and the order of the steps may be set in various ways, and the steps may be performed almost simultaneously.

Referring to FIG. 2 , the method for generating a disease prediction model is a method for generating a disease prediction model using a device for generating a disease prediction model. S130). Here, the apparatus for generating a disease prediction model may be the above-described apparatus for generating a disease prediction model ( 100 of FIG. 1 ).

The training data generation step S100 is a step in which the disease prediction model generating apparatus pre-processes training data including medical information of each of a plurality of insurers to generate a training data set. In the AI model testing step ( S120 ), the disease prediction model generating apparatus performs performance evaluation including accuracy evaluation for each of a plurality of AI models based on test data including some of the training data included in the training data set. It is a step to select an AI model with the highest performance index among a plurality of AI models by performing In the artificial intelligence model generation step (S130), the disease prediction model generation device sets the personal data, medical treatment data, and lifestyle data of insurers included in the learning data set as input variables, and disease occurrence of the insurers included in the learning data set. It is a step of generating a user-customized disease prediction model by setting the data as an output variable and learning the selected AI model according to the AI model test step (S120) based on the training data set.

Referring to FIG. 3 , the training data generation step S100 may include a training data reception step S111 , a training data integration and pre-processing step S112 , and a training data set generation step S113 .

The learning data receiving step S111 is a step in which the disease prediction model generating apparatus receives learning data including medical information from the National Health Insurance Corporation (KNHIS). The medical information may include insurance eligibility data, insurance premium data, birth data, death data, medical treatment data, disease history data, and health checkup data of each of the plurality of insurers. For example, the medical information may be in the form of an Excel file classified into several.

The learning data integration and pre-processing step ( S112 ) is a step in which the disease prediction model generating apparatus integrates the training data into one and pre-processes the integrated training data to generate a training data set. In the learning data integration and pre-processing step (S112), the disease prediction model generating device pre-processes the learning data, from medical information to the labeling target of the training data set for the main and injury data, injury data, personal data, lifestyle data, and medical treatment data. It may include the step of setting it to data. The learning data integration and pre-processing step (S112) is a step of extracting and integrating each of the plurality of insurers' personal injury data, injury data, personal data, lifestyle data and medical treatment data from the learning data in which the disease prediction model generating device is integrated. may include. The learning data integration and pre-processing step ( S112 ) may include a step in which the disease prediction model generating device integrates the main-injured disease data, the wounded disease data, personal data, lifestyle data, and medical treatment data through a common key. Learning data integration and The pre-processing step ( S112 ) may include analyzing the correlation between the integrated data to determine data importance and data reliability. The main-injury data includes information on the health and illness that insurers have the greatest demand for treatment or testing. Injury data includes information on injuries and diseases that have affected patient care as a condition that occurred with or occurred during the insurers' treatment period. Personal data includes the gender and age of the insurer. The lifestyle data includes information on the amount of smoking, the number of days smoked, the number of days drunk, and the amount of alcohol consumed. Medical data includes at least one or more of the insurer's medical treatment period and treatment date, past disease history, recent disease history and type of examination, drug treatment history, visiting department, family history, hepatitis B antigen holding and current drug use do. The common key may be the insurers' personal unique number (RN_INDI). The correlation between the integrated data is analyzed through Cramer V test, Random Forest and Linear Regression. Cramer V test and Random Forest can identify the importance of data in the integrated data. Linear Regression can determine the reliability of data from the integrated data. A method of analyzing the correlation between the integrated data will be described in detail with reference to FIGS. 6 to 8 to be described later.

The training data set generation step S113 is a step in which the disease prediction model generating apparatus recognizes errors in the learning data that are correlated and integrated, and generates a training data set by correcting the errors. The error correction of the training data may include at least one of missing value processing, outlier processing, data augmentation, and data unit conversion. Missing value processing can be corrected by removing the missing value from the training data or inserting a mode value. Data augmentation can be corrected by identifying Imbalanced Data of training data and applying Synthetic Minority Oversampling Technique (SMOTE). Data unit conversion is to fit data units by converting all data within a specific range through MinMax Scaling. For example, data unit conversion may be performed after data augmentation through SMOTE. More specifically, data imbalance can be resolved through MLSMOTE_MLP, which combines MLSMOTE, a type of data imbalance processing method by the disease prediction model generation chapter, and MLP (Multi-Layer Perceptron). MLSMOTE_MLP calculates IRPL (Imbalance ratio per label, individually calculated value for each label) for the data with the number of labels |L| and the number of labels |N| through Equation (1) below, and MIR ( Mean Imbalance ratio, IRPL average value) is calculated. And, as shown in Equation (3) below, if IRPL(I) > MIR, it is classified as a tail label (unbalanced minority data). The tail label is classified as the median rather than the mean in consideration of data distortion and specificity. After classifying the tail label, we proceed to augment the feature vector data suitable for the tail label. Data augmentation may be using NN Euclidean distance. The augmented data is input to the MLP model to finally correct the data imbalance. In order to improve data imbalance correction performance of the MLP model, the apparatus for generating a disease prediction model may tune a hyperparameter.

Formula (1)

Equation (2)

Equation (3)

In addition, in an embodiment of the present invention, data imbalance can be processed by using SMOTE_MLP, which is a combination of SMOTE and MLP, which is a type of data imbalance processing method. The main feature of SMOTE_MLP is to convert multi-label to multi-class, and may include the order of SMOTE, ML/DL, and model tuning.

An example of converting a multilabel to a multiclass is as follows.

(A,B) -> 1,

(A,E) -> 2,

(C,E) -> 3,

(D,E) -> 4,

(A,B) -> 1,

(A,B) -> 1,

(D,E) -> 4

Referring to FIG. 4 , the AI model testing step S120 may include a test data generation step S121 , an AI model performance evaluation step S122 , and an AI model selection step S123 .

The test data generation step S121 is a step in which the disease prediction model generating apparatus generates some of the training data included in the training data set as test data. The test data includes test input data and test result data such as personal data, medical treatment data, and lifestyle data of any one of the plurality of insurers.

In the AI model performance evaluation step (S122), the disease prediction model generating device performs performance evaluation on each of a plurality of AI models based on the test data to derive a performance index for each of the plurality of AI models. is a step The plurality of AI models may include two or more of Random Forest, Decision Tree, XGBoost, Naive Bayes, and DNN. Performance evaluation includes accuracy evaluation, precision evaluation, recall evaluation, and F1 score evaluation for each of a plurality of artificial intelligence models.

The artificial intelligence model selection step S123 is a step in which the disease prediction model generating apparatus selects an artificial intelligence model having the highest performance index among a plurality of artificial intelligence models. More specifically, it is a step of selecting an artificial intelligence model having the highest F1 score in which an accuracy index, a precision index, a recall index, and a reliability are equally calculated among a plurality of artificial intelligence models. The AI model selection can be selected by the disease prediction model generation device using a decision tree.

Referring to FIG. 5 , the artificial intelligence model generation step S130 may include an input variable and output variable setting step S131 and a disease prediction model generation step 132 .

In the step S131 of setting input variables and output variables, the disease prediction model generating apparatus sets personal data, medical treatment data, and lifestyle data of insurers included in the learning data set as input variables, and the insurers included in the learning data set. This is the step of setting their disease occurrence data as an output variable. The input variable and the output variable are selected by referring to the data importance and reliability derived through the learning data correlation analysis integrated in the learning data integration and pre-processing step (S112). The input variables selected by referring to the correlation analysis between the integrated learning data include personal data, medical treatment data, and lifestyle data of insurers. The personal data includes at least one of age, gender, and date of birth. Personal data must include age and gender data. Medical data includes at least one or more of the insurer's medical treatment period and date, past disease history, recent disease history, type of examination, visiting department, family history, drug treatment history, hepatitis B antigen holding and current drug use do. Medical data must include historical disease history, recent disease history, visiting department and drug treatment history data. The lifestyle data includes at least one of smoking status, smoking period, smoking amount, drinking status, drinking period, and drinking amount. Lifestyle data must include data on whether or not smoking, smoking period, smoking amount, drinking status, drinking period and drinking amount data. The disease occurrence data includes cancer occurrence data, heart disease occurrence data, brain disease occurrence data, liver disease occurrence data, diabetes occurrence data, hypertension occurrence data, thyroid disease occurrence data, renal gland occurrence data, glaucoma occurrence data, and arteriosclerosis occurrence data. may include. The output variable may be set to include at least one or more data among data included in the disease occurrence data.

In the disease prediction model generation step 132 , the disease prediction model generating device learns the AI model selected according to the AI model performance evaluation based on the training data set including the input variable and the output variable, and the user-customized disease prediction model is the step to create The correlation between the main and injured disease data of the training data set and personal data, lifestyle data, and medical treatment data has the highest correlation with the output variable of the user-customized disease prediction model. For example, insurer A's main disease is lung cancer, and the injured disease is bronchial cancer. At this time, insurer A suffered from chronic cough disease within 1 year and smoked 5 days a week. When an input variable having the above treatment data and lifestyle data is input to a user-customized disease prediction model, the user-customized disease prediction model derives lung cancer and bronchial cancer as output variables (disease prediction information), and converts the output variables into cancer Categorize by occurrence data.

6 to 8 are diagrams illustrating the results of analyzing the correlation between the learning data integrated in the learning data integration and pre-processing step (S112). Hereinafter, a correlation analysis result between the integrated learning data will be described with reference to FIGS. 6 and 8 .

6 is a diagram illustrating a correlation analysis result between integrated learning data using Cramer's V test.

The disease prediction model generating apparatus classifies the integrated learning data into <Table 1> and the data.

<Table 1> Classification of integrated training data classification code Code Description Age age Q_DRK_AMT_V09N 1 drink Q_DRK_FRQ_V09N Weekly drinking week Q_SMK_NOW_AMT_V09N (Current) Smoking amount per day Q_SMK_NOW_DRT (Current) Smoking Period Q_SMK_PRE_AMT (Past) Smoking amount per day Q_SMK_PRE_DRT (Past) Smoking period Q_SMK_YN smoking status Q_HBV_AG Hepatitis B antigen carrier Q_FHX_ETC (Family history) Other (including cancer) Q_FHX_DM (Family history) Diabetes mellitus Q_FHX_HTN (Family history) High blood pressure Q_FHX_HTDZ (Family history) Heart disease (myocardial infarction angina) Q_PHX_TX_STK (Self) Stroke (Stroke) Whether drug treatment Q_PHX_DX_ETC (Individual) Past history of other (including cancer) diseases Q_PHX_DX_PTB (Self) Past history of pulmonary tuberculosis Q_PHX_TX_DLD (Self) Hyperlipidemia drug treatment Q_PHX_DX_DM (You) Have a past history of diabetes Q_PHX_DX_HTN (Self) Do you have a past history of high blood pressure? Q_PHX_DX_HTDZ (Self) Do you have a past history of heart disease? Q_PHX_DX_STK (You) Have a history of stroke SEX gender DETAIL_TMSG_SUBJ_CD Detailed Specialized Course Code MCEX_SICK_SYM The morbidity that the patient received treatment during the treatment period SICK_SYM1 columnar Q_PHX_TX_HTN Whether (the person) is treated with medication for high blood pressure Q_PHX_DX_DLD Past history of hyperlipidemia (dyslipidemia) Q_PHX_TX_ETC Whether (person) other (including cancer) drug treatment Q_FHX_STK (Family history) Stroke (stroke)

When the disease prediction model generating device inputs the classified data as shown in <Table 1> into Cramer’s V test, the results shown in <Table 2>, <Table 3> and <Table 4> are derived.

<Table 2> Correlation of medical course code data

<Table 3> Correlation of drug ingredient name extraction data

<Table 4> About the target value of the medical course code data
correlation

Referring to 6, the result graphs of <Table 2>, <Table 3> and <Table 4> above are shown as follows. Here, data with a Cramer's V test value of 0.5 or higher is judged by the disease prediction model generating device as meaningful data that can be used as an artificial intelligence model learning variable because it has a high correlation with the main and wounded disease data and the wounded disease data. Significant data may be, for example, MCEX_SICK_SYM (injured) and SICK_SYM1 (injured). Data with Cramer's V test values of 0.3 or more and less than 0.5 have low correlation with the main and wounded soldiers data, but are judged to be usable as candidate variables for artificial intelligence model learning. The candidate variable data may be, for example, Q_PHX_DX_STK (whether or not (the person) has a past history of stroke) and Q_PHX_TX_STK (whether (the person) is treated with drugs for stroke (paralysis)).

7 is a diagram illustrating a correlation analysis result between learning data integrated through a random forest.

Referring to FIG. 7 , the disease prediction model generating device inputs the classified data as shown in Table 1 above from the integrated training data into the random forest model. The random forest model analyzes the importance of the input data. The random forest model derives the results shown in <Table 5>, <Table 6>, <Table 7>, <Table 8> and <Table 9> by analyzing the input data.

<Table 5> RF Ranking of Medical Course Code (Target) Data

<Table 6> RF Ranking of Medical Course Code (Public Disease) Data

<Table 7> Medical course code (injured disease) data
RF Ranking

<Table 8> Extraction data of drug ingredient name
RF Ranking

<Table 9> Drug ingredient name (injured disease) extraction data
RF Ranking

Among the results of <Table 5> to <Table 9>, when the data having the highest value is expressed as a graph, it is expressed as shown in FIG. 7 . Here, data having a random forest model result value of more than 100,000 may be determined by the disease prediction model generating device as having very high data importance. Data whose random forest model result value is greater than 10,000 and less than or equal to 100,000 may be, for example, SICK_SYM1 (injured disease), MCEX_SICK_SYM (injured disease), DETAIL_TMSG_SUBJ_CD (subspecialty code) and age (age) data. Data with a random forest model result value of more than 10,000 and less than 100,000 are also judged to be of high importance. Data with a random forest model result value greater than 10,000 and less than or equal to 100,000 are, for example, SEX (gender), Q_DRK_FRQ_V09N (weekly drinking weeks), (Q_HBV_AG (hepatitis B antigen holder), Q_DRK_AMT_V09N (amount of alcohol consumed), Q_PHX_TX_STK(( Self) Stroke (paralysis) medication treatment), Q_SMK_PRE_DRT((past) smoking period), Q_SMK_NOW_DRT((current) smoking period), Q_PHX_DX_HTN((person) past history of hypertension), Q_PHX_TX_HTN((person) high blood pressure medication treatment) , Q_DRK_AMT_V09N (amount of alcohol per drink).

8 is a diagram illustrating a correlation analysis result between learning data integrated through linear regression.

Referring to FIG. 8 , the disease prediction model generating device inputs the classified data as shown in Table 1 from the integrated training data into the linear regression model. The linear regression model analyzes the reliability of the input data. The linear regression model derives the results shown in <Table 10>, <Table 11>, <Table 12> and <Table 13> by analyzing the input data.

<Table 10> Medical course code data P-value

<Table 11> P-value of medical course code extraction data

<Table 12> About the target value of the medical course code
P-value

<Table 13> Drug ingredient name extraction data
P-value

When the result value (P-value) of <Table 10> to <Table 13> is expressed as a graph, it is expressed as in FIG. 8 . Here, data having a linear regression model result value (P-value) of 0 or more and less than 0.5 is determined to have high data reliability. Data with a random forest model result value of 0 or more and less than 0.5 are, for example, SICK_SYM1 (main disease), MCEX_SICK_SYM (corresponding disease treated by the patient during treatment), DETAIL_TMSG_SUBJ_CD (subspecialty code), SEX (gender), Q_PHX_DX_HTN ((You) have a history of high blood pressure, Q_PHX_DX_DLD (Have a history of hyperlipidemia (dyslipidemia)), Q_PHX_TX_DM((You) have diabetes drug treatment), Q_PHX_TX_DLD((You) have been treated with hyperlipidemia medication), Q_PHX_TX_ETC((You) Other (including cancer) drug treatment), Q_HBV_AG (Hepatitis B antigen holder), Q_SMK_PRE_DRT ((Past) smoking period), Q_SMK_PRE_AMT((Past) Amount smoked per day), Q_DRK_FRQ_V09N (Number of drinks per week), Q_DRK_AMT_V09N (Amount drunk per day) and age. Data with a random forest model result value less than 0.001 are, for example, Q_FHX_STK ((family history) stroke (stroke)), Q_SMK_YN (smoking status), Q_SMK_NOW_DRT ((current) smoking period), and Q_SMK_NOW_AMT_V09N ((current) daily smoking amount) days can Data with a random forest model result value of less than 0.05 are Q_PHX_TX_HTN((your own) hypertension drug treatment), Q_FHX_HTN((family history) high blood pressure status), Q_PHX_DX_STK((your own) history of stroke), Q_PHX_DX_DM((your own) diabetes mellitus or not ), Q_PHX_TX_STK (whether (the person) is treated with drugs for stroke (paralysis)), and Q_FHX_HTDZ (whether (family history) heart disease (myocardial infarction/angina)).

9 is a diagram illustrating a terminal 1000 communicatively connected to a user-customized disease prediction system (hereinafter, referred to as a “disease prediction system 900”) according to another embodiment of the present invention.

Referring to FIG. 9 , the disease prediction system 900 and the terminal 1000 may be interconnected through a wired or wireless communication network. The disease prediction system 900 may be formed of a cloud computing server such as Software as a Service (SaaS), Platform as a Service (PaaS), or Infrastructure as a Service (IaaS). Terminal 1000 is, for example, a web browser (WEB Browser) is mounted laptop, desktop (desktop), laptop (laptop), a wireless communication device or smart phone that guarantees portability and mobility, and a tablet PC including a touchpad It may refer to all types of handheld-based wireless communication devices such as, for example. A communication network is a wired network such as a Local Area Network (LAN), a Wide Area Network (WAN) or a Value Added Network (VAN), or any network such as a mobile radio communication network or a satellite network. It can be implemented as a kind of wireless network.

The terminal 1000 may transmit user input information to the disease prediction system 900 , and may output the disease prediction information received from the disease prediction system 900 and display it on a display.

The disease prediction system 900 and the terminal 1000 will be described in more detail below with reference to FIGS. 10 and 11 .

10 is a block diagram illustrating the configuration of a user-customized disease prediction system 900 .

Referring to FIG. 10 , the disease prediction system 900 includes a communication module 910 , a memory 920 , a processor 940 , and may further include a database 930 .

The communication module 910 may transmit/receive information to and from the terminal 1000 . The communication module 910 may include a device including hardware and software necessary for transmitting and receiving signals such as control signals or data signals through wired/wireless connection with other network devices.

The memory 920 stores a program for providing user-customized disease prediction information. The name of the user-customized disease prediction information providing program is set for convenience of explanation, and the name itself does not limit the function of the program. The memory 920 may be configured to store at least one of data input to the communication module 910 , data required for a function performed by the processor 940 , and data generated according to execution by the processor 940 . . The memory 920 should be interpreted as a generic term for a non-volatile storage device that continuously maintains stored information even when power is not supplied, and a volatile storage device that requires power to maintain the stored information. The memory 920 may perform a function of temporarily or permanently storing data processed by the processor 940 . The memory 920 may include magnetic storage media or flash storage media in addition to a volatile storage device that requires power to maintain stored information, but the scope of the present invention is not limited thereto. not.

The database 930 may be a place in which data required for generating user-customized disease prediction information is stored. The database 930 may be built in a partial area of the memory 920 or implemented as separate hardware.

The processor 940 is configured to execute a program for providing user-customized disease prediction information stored in the memory 920 . The processor 940 may include various types of devices for controlling and processing data. The processor 940 may refer to a data processing device embedded in hardware having a physically structured circuit to perform a function expressed as a code or an instruction included in a program. In one example, the processor 940 includes a microprocessor, a central processing unit (CPU), a processor core, a multiprocessor, an application-specific integrated circuit (ASIC), an FPGA (FPGA). field programmable gate array), but the scope of the present invention is not limited thereto.

The processor 940 is configured to execute a user-customized disease prediction information providing program to perform the following functions and procedures.

The processor 940 receives user input information from the terminal 1000 through the communication module 910 . The user input information includes the user's personal data, the user's medical treatment data, and lifestyle data. The user's personal data includes at least one or more of a name, age, date of birth, and gender. Users' personal data must include age and gender data. The user's medical treatment data includes at least one or more of the user's medical treatment date, examination type, past medical history data, recent medical history data, recent department data, family history, and drug treatment history data. The user's medical treatment data necessarily includes past medical history data, recent medical history data, recent department data, and drug treatment history data. The user's lifestyle data may include at least one of drinking data, smoking data, hepatitis B antigen holding data, and current drug intake data. The user's lifestyle data necessarily includes drinking data and smoking data. The drinking data includes information such as the drinking period, whether or not drinking, and the amount of drinking, and the smoking data includes information such as whether or not smoking, the period of smoking and the amount of smoking.

The processor 940 generates user disease prediction information based on user input information using the disease prediction model. The disease prediction model is generated by the processor 940 through the following functions and procedures. The disease prediction model is based on the learning data set including personal data, medical treatment data, lifestyle data, and disease occurrence data of the insurers received from the National Health Insurance Corporation server through the processor 940, personal data, medical treatment data and It is an artificial intelligence model learned by using lifestyle data as an input variable and the disease occurrence data as an output variable. The personal data includes at least one of age, gender, and date of birth. Personal data must include age and gender data. Medical data includes at least one or more of the insurer's medical treatment period and date, past disease history, recent disease history, type of examination, visiting department, family history, drug treatment history, hepatitis B antigen holding and current drug use do. Medical data must include historical disease history, recent disease history, visiting department and drug treatment history data. The lifestyle data includes at least one of smoking status, smoking period, smoking amount, drinking status, drinking period, and drinking amount. The lifestyle data includes at least one of smoking status, smoking period, smoking amount, drinking status, drinking period, and drinking amount. Lifestyle data must include data on whether or not smoking, smoking period, smoking amount, drinking status, drinking period and drinking amount data. The disease occurrence data includes cancer occurrence data, heart disease occurrence data, brain disease occurrence data, liver disease occurrence data, diabetes occurrence data, hypertension occurrence data, thyroid disease occurrence data, renal gland occurrence data, glaucoma occurrence data, and arteriosclerosis occurrence data. may include. The output variable may be set by the processor 940 to include at least one or more of data included in the disease occurrence data. The processor 940 generates a disease prediction model by learning an artificial intelligence model based on a training data set including an input variable and an output variable. In more detail, when the disease prediction model is generated, the processor 940 collects personal data, injury data, personal data, lifestyle data, and medical treatment data from learning data including data of insurers received from the National Health Insurance Corporation server. It can be set as feature data that is a labeling target of the training data set. The processor 940 may determine data importance and reliability by performing correlation analysis of data set by the disease prediction model as feature data in the training data set, and may set input variables and output variables among data having high importance and reliability. let it be For example, insurer A's main disease is lung cancer, and the injured disease is bronchial cancer. At this time, insurer A suffered from chronic cough disease within 1 year and smoked 5 days a week. When an input variable having the above treatment data and lifestyle data is input to a user-customized disease prediction model, the user-customized disease prediction model outputs lung cancer and bronchial cancer as output variables (disease prediction information), and outputs the output variables to cancer It can be categorized by occurrence data. The correlation between the feature data performed by the processor 940 may be identified by analyzing through Cramer V test, Random Forest, and Linear Regression.

The processor 940 provides the user disease prediction information to the terminal 1000 . The user disease prediction information includes dangerous disease prediction information related to at least one or more data among data included in the insurers' disease occurrence data, self-diagnosis method information corresponding to the dangerous disease prediction information, and prevention method information corresponding to the dangerous disease prediction information. include When the user disease prediction information includes a plurality of dangerous disease prediction information related to two or more data among the data included in the disease occurrence data, the processor 940 classifies the plurality of risk disease prediction information into a high-risk group and a low-risk group Thus, it can be provided to the terminal 1000 .

The processor 940 may further include providing insurance product recommendation information corresponding to the user disease prediction information to the terminal 1000 . The insurance product recommendation information may include an insurance product name and insurance product terms corresponding to the user's disease prediction information. The processor 940 may receive insurance product data from an insurance company server or terminal. The insurance product data includes insurance product names and insurance product terms and conditions data. Insurance policy data may include information such as disease classification numbers and disease names for insured diseases. The insurance product data may be a PDF file such as an insurance product policy. The processor 940 extracts the disease name and disease classification number from the insurance product data, and maps the extracted data and user disease prediction information to obtain insurance product recommendation information for the disease classification number and disease name corresponding to the user disease prediction information. create Then, the generated insurance product recommendation information is provided to the terminal 1000 . For example, when the dangerous disease information of the user's disease prediction information is a malignant neoplasm of the digestive system, an insurance product corresponding to the dangerous disease information may be recommended. When extracting disease names and disease classification numbers from insurance product data. This can be done using Python.

11 is a block diagram illustrating the configuration of the terminal 1000 .

Referring to FIG. 11 , the terminal 1000 may include a communication module 1010 , a memory 1020 , an input/output module 1030 , and a processor 1040 .

The communication module 1010 transmits/receives information to and from the disease prediction system 900 . The communication module 910 may include a device including hardware and software necessary for transmitting and receiving signals such as control signals or data signals through wired/wireless connection with other network devices.

The memory 1020 stores a user-customized disease prediction information recommendation program. The name of the user-customized disease prediction information recommendation program is set for convenience of explanation, and the name itself does not limit the function of the program. The memory 1020 stores at least any one or more of information and data input to the communication module 1010 , information and data necessary for a function performed by the processor 1040 , and data generated according to the execution of the processor 1040 . can The memory 1020 should be interpreted as a generic term for a non-volatile storage device that continuously maintains stored information even when power is not supplied, and a volatile storage device that requires power to maintain the stored information. Also, the memory 1020 may perform a function of temporarily or permanently storing data processed by the processor 1040 . The memory 1020 may include magnetic storage media or flash storage media in addition to a volatile storage device that requires power to maintain stored information, but the scope of the present invention is not limited thereto. not.

The input/output module 1030 may receive information, data, etc. transmitted from the outside to the terminal 1000 , or may output information and data possessed by the terminal 1000 to the outside. For example, the input/output module 1030 may include a display, a touchpad, a speaker, and a microphone.

The processor 1040 may include various types of devices for controlling and processing data. The processor 1040 may refer to a data processing device embedded in hardware having a physically structured circuit to perform a function expressed as a code or an instruction included in a program. In one example, the processor 1040 is a microprocessor, a central processing unit (CPU), a processor core, a multiprocessor, an application-specific integrated circuit (ASIC), an FPGA ( field programmable gate array), but the scope of the present invention is not limited thereto. The processor 1040 is configured to execute a user-customized disease prediction information recommendation program (hereinafter, referred to as a “disease prediction recommendation program”) stored in the memory 1020 to perform the following functions and procedures.

Information transmitted from the processor 1040 to the disease prediction system 900 may be information input to various interfaces output through the input/output module 1030 . The interface may be received from the disease prediction system 900 or generated by the processor 1040 .

The processor 1040 generates a user information input interface in the input/output module 1030 , and transmits the user input information input to the interface to the user disease prediction system 900 through the communication module 1010 . Since the user input information is the same as the user input information described with reference to FIG. 10, a description thereof will be omitted.

The processor 1040 may receive user disease prediction information from the disease prediction system 900 and output it to the input/output module 1030 . The user disease prediction information may include dangerous disease information, a self-diagnosis method corresponding to the dangerous disease information, and prevention method information corresponding to the dangerous disease information.

The processor 1040 may receive insurance product recommendation information corresponding to the user's disease prediction information from the disease prediction system 900 and output the received insurance product recommendation information to the input/output module 1030 . The insurance product recommendation information may include an insurance product name and insurance product terms corresponding to the user's disease prediction information.

12 is a flowchart illustrating a user-customized disease prediction method for providing customized disease prediction information to a user according to another embodiment of the present invention, and FIG. 13 is an additional step of the user-customized disease prediction method shown in FIG. It is a drawing shown for explanation. Hereinafter, a user-customized disease prediction method will be described with reference to FIGS. 12 and 13 . Each step of the user-customized disease prediction method to be described below may be performed by the user-customized disease prediction system 900 described above with reference to FIGS. 9 to 11 . Accordingly, the contents of another embodiment of the present invention described above with reference to FIGS. 9 to 11 can be equally applied to the embodiments of FIGS. 12 and 13 to be described below, and the descriptions described above with reference to FIGS. 9 to 11 . Contents that overlap with the above should be omitted. The steps described in FIGS. 12 and 13 are not necessarily performed in order, and the order of the steps may be set in various ways, and each step may be performed almost simultaneously.

Referring to FIG. 12 , the user-customized disease prediction method is a method of providing customized disease prediction information to a user through a communication connection between a server and a terminal. ) and transmitting user disease prediction information ( 930 ). Here, the server and the terminal may be a customized disease prediction system ( 900 in FIG. 9 ) and a terminal ( 1000 in FIG. 9 ) shown in FIG. 9 , respectively.

The user input information receiving step S1210 is a step in which the server receives user input information from the terminal. The user input information includes the user's personal data, the user's medical treatment data, and lifestyle data. The user's personal data includes at least one or more of a name, age, date of birth, and gender. Users' personal data must include age and gender data. The user's medical treatment data includes at least one or more of the user's medical treatment date, examination type, past medical history data, recent medical history data, recent department data, family history, and drug treatment history data. The user's medical treatment data necessarily includes past medical history data, recent medical history data, recent department data, and drug treatment history data. The user's lifestyle data may include at least one of drinking data, smoking data, hepatitis B antigen holding data, and current drug intake data. The user's lifestyle data necessarily includes drinking data and smoking data. The drinking data includes information such as the drinking period, whether or not drinking, and the amount of drinking, and the smoking data includes information such as whether or not smoking, the period of smoking and the amount of smoking.

The user disease prediction information generation step ( S1220 ) is a step in which the server generates user disease prediction information based on user input information received from the terminal using the disease prediction model. The disease prediction model is generated by the server. The server uses personal data, medical treatment data and lifestyle data as input variables based on a learning data set including personal data, medical treatment data, lifestyle data, and disease occurrence data of insurers received from the National Health Insurance Corporation server, and A trained disease prediction model is generated using the disease occurrence data as an output variable. The personal data includes at least one of age, gender, and date of birth. Personal data must include age and gender data. Medical data includes at least one or more of the insurer's medical treatment period and date, past disease history, recent disease history, type of examination, visiting department, family history, drug treatment history, hepatitis B antigen holding and current drug use do. Medical data must include historical disease history, recent disease history, visiting department and drug treatment history data. The lifestyle data includes at least one of smoking status, smoking period, smoking amount, drinking status, drinking period, and drinking amount. Lifestyle data must include data on whether or not smoking, smoking period, smoking amount, drinking status, drinking period and drinking amount data. The disease occurrence data includes cancer occurrence data, heart disease occurrence data, brain disease occurrence data, liver disease occurrence data, diabetes occurrence data, hypertension occurrence data, thyroid disease occurrence data, renal gland occurrence data, glaucoma occurrence data, and arteriosclerosis occurrence data. may include. The output variable may be set by the server to include at least one or more data among data included in the disease occurrence data. In this way, the server generates a disease prediction model by learning the artificial intelligence model based on the training data set including the input variable and the output variable. In more detail, when the server generates a disease prediction model, the training data set includes personal injury data, injury data, personal data, lifestyle data and medical treatment data from the learning data including data of insurers received from the National Health Insurance Corporation server. It can be set as the feature data that is the labeling target of The server performs correlation analysis of data set by the disease prediction model as feature data in the training data set to determine data importance and reliability, and enables input and output variables to be set among data with high importance and reliability. For example, insurer A's main disease is lung cancer, and the injured disease is bronchial cancer. At this time, insurer A suffered from chronic cough disease within 1 year and smoked 5 days a week. When an input variable having the above treatment data and lifestyle data is input to a user-customized disease prediction model, the user-customized disease prediction model outputs lung cancer and bronchial cancer as output variables (disease prediction information), and outputs the output variables to cancer It can be categorized by occurrence data.

The user disease prediction information transmission step ( S1230 ) is a step in which the server provides the user disease prediction information to the terminal. The user disease prediction information includes dangerous disease prediction information related to at least one or more data among data included in the insurers' disease occurrence data, self-diagnosis method information corresponding to the dangerous disease prediction information, and prevention method information corresponding to the dangerous disease prediction information. include When the user disease prediction information includes a plurality of dangerous disease prediction information related to two or more data among the data included in the disease occurrence data, the server classifies the plurality of dangerous disease prediction information into a high-risk group and a low-risk group and provides it to the terminal. can

FIG. 13 is a diagram illustrating an additional step of the method for predicting a user-customized disease shown in FIG. 12 .

Referring to FIG. 13 , the user-customized disease prediction method may further include providing insurance product recommendation information. The step of providing insurance product recommendation information may include an insurance product data receiving step (S1310) and an insurance product recommendation information generating step (S1320).

The insurance product data receiving step S1310 is a step in which the server receives insurance product data from the insurance company server or terminal. The insurance product data includes insurance product names and insurance product terms and conditions data. Insurance policy data may include information such as disease classification numbers and disease names for insured diseases. The insurance product data may be a PDF file such as an insurance product policy.

The insurance product recommendation information generation step ( S1320 ) is a step in which the server generates insurance product recommendation information corresponding to the user's disease prediction information and provides it to the terminal. More specifically, the server extracts the disease name and disease classification number from the insurance product data, maps the extracted data and the user disease prediction information, and recommends the insurance product for the disease classification number and disease name corresponding to the user disease prediction information create information In addition, the generated insurance product recommendation information may be provided to the terminal. The insurance product recommendation information may include an insurance product name and insurance product terms corresponding to the user's disease prediction information. As above, for example, when the dangerous disease information of the user's disease prediction information derived from the server is a malignant neoplasm of the digestive system, an insurance product corresponding to the dangerous disease information may be recommended.

The disease prediction model generation method and the user disease prediction method using the same according to the embodiments of the present invention described above are also in the form of a recording medium including instructions executable by a computer, such as a program module executed by a computer. can be implemented. Computer-readable media can be any available media that can be accessed by a computer and includes both volatile and nonvolatile media, removable and non-removable media. Also, computer-readable media may include computer storage media. Computer storage media includes both volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data.

Those of ordinary skill in the art to which the present invention pertains will be able to understand that it can be easily modified into other specific forms without changing the technical spirit or essential features of the present invention based on the above description. Therefore, it should be understood that the embodiments described above are illustrative in all respects and not restrictive. The scope of the present invention is indicated by the following claims, and all changes or modifications derived from the meaning and scope of the claims and their equivalents should be construed as being included in the scope of the present invention. The scope of the present application is indicated by the following claims rather than the above detailed description, and all changes or modifications derived from the meaning and scope of the claims and their equivalent concepts should be construed as being included in the scope of the present application.

100: device for generating a disease prediction model
110: memory
120: communication module
130: processor
900: User-customized disease prediction system
910: communication module
920: memory
930: database
940: processor
1000: terminal
1010: communication module
1020: memory
1030: input/output module
1040: Processor

Claims (20)

A method of generating an artificial intelligence model that provides customized disease prediction information to a user by using a device for generating a user-customized disease prediction model, the method comprising:
a) the device. generating a training data set by pre-processing training data including medical information of each of a plurality of insurers;
b) the device performs performance evaluation including accuracy evaluation for each of a plurality of AI models based on test data including some of the training data included in the training data set, and the plurality of AI models selecting an AI model having the highest performance index among them; and
c) setting the personal data, medical treatment data, and lifestyle data of the insurers included in the learning data set as input variables, and setting the disease occurrence data of the insurers included in the learning data set as output variables, as an output variable, the learning data Generating a user-customized disease prediction model by learning the artificial intelligence model selected according to step b) based on the set,
The performance evaluation includes at least one of an accuracy evaluation, a precision evaluation, a recall evaluation, and an F1 score evaluation for each of the plurality of artificial intelligence models, and
Wherein the performance index includes at least one of an accuracy index, a precision index, a recall index, and an F1 score index corresponding to the performance evaluation. The method of claim 1,
Wherein the medical information includes insurance eligibility data, insurance premium data, birth data, death data, medical treatment data, disease history data, and health checkup data of each of the plurality of insurers. 3. The method of claim 2,
Step a) is,
Disease prediction model comprising the step of, by the device, setting, among the data included in the medical information, main-injury disease data, injured disease data, personal data, lifestyle data, and medical treatment data as characteristic data to be labeled in a learning data set creation method. delete The method of claim 1,
The disease occurrence data includes cancer occurrence data, heart disease occurrence data, brain disease occurrence data, liver disease occurrence data, diabetes occurrence data, hypertension occurrence data, thyroid disease occurrence data, renal gland occurrence data, glaucoma occurrence data, and arteriosclerosis occurrence data. contains data;
The method of generating a disease prediction model, wherein the output variable is set to include at least one or more data among data included in the disease occurrence data. In the device for generating a user-customized disease prediction model,
a memory for storing a disease prediction model generation program; and
A processor for executing the program stored in the memory;
The processor, by executing the disease prediction model generation program,
A training data set is generated by preprocessing training data including medical information of each of a plurality of insurers, and each of a plurality of artificial intelligence models is based on test data including some of the training data included in the training data set. Selects an AI model having the highest performance index among the plurality of AI models by performing performance evaluation including accuracy evaluation for A user-customized disease prediction model by setting lifestyle data as an input variable and setting the disease occurrence data of the insurers included in the learning data set as an output variable to learn the selected artificial intelligence model based on the learning data set create,
The performance evaluation includes at least one of an accuracy evaluation, a precision evaluation, a recall evaluation, and an F1 score evaluation for each of the plurality of artificial intelligence models, and
Wherein the performance index includes at least one of an accuracy index, a precision index, a recall index, and an F1 score index corresponding to the performance evaluation, the disease prediction model generating apparatus. 7. The method of claim 6,
The medical information includes insurance eligibility data, insurance premium data, birth data, death data, medical treatment data, disease history data, and health checkup data of each of the plurality of insurers,
The processor, by executing the disease prediction model generation program,
In the pre-processing of the learning data, setting the column and injury data, injury data, personal data, lifestyle data, and medical treatment data in the medical information as feature data to be labeled in the learning data set is further performed, disease prediction model generation Device. delete 7. The method of claim 6,
The disease occurrence data includes cancer occurrence data, heart disease occurrence data, brain disease occurrence data, liver disease occurrence data, diabetes occurrence data, hypertension occurrence data, thyroid disease occurrence data, renal gland occurrence data, glaucoma occurrence data, and arteriosclerosis occurrence data. contains data;
The apparatus for generating a disease prediction model, wherein the output variable is set to include at least one or more data among data included in the disease occurrence data. ◈Claim 10 was abandoned when paying the registration fee.◈ A method of providing customized disease prediction information to a user through a communication connection between a terminal and a server,
i) receiving, by the server, user input information from the terminal;
ii) the server collects the personal data, medical treatment data and lifestyle data based on the learning data set including personal data, medical treatment data, lifestyle data and disease occurrence data of the insurers received from the National Health Insurance Corporation server generating user disease prediction information based on the user input information using a disease prediction model learned using the disease occurrence data as an input variable and the disease occurrence data as an output variable; and
iii) the server, comprising the step of providing the user disease prediction information to the terminal,
The disease prediction model performs performance evaluation including accuracy evaluation on each of a plurality of artificial intelligence models based on test data including some of the data included in the training data set, and among the plurality of artificial intelligence models An artificial intelligence model having the highest performance index is selected, and the selected artificial intelligence model is learned based on the input variable and the output variable,
The performance evaluation includes at least one of an accuracy evaluation, a precision evaluation, a recall evaluation, and an F1 score evaluation for each of the plurality of artificial intelligence models, and the performance indicator includes an accuracy indicator corresponding to the performance evaluation, a precision indicator, A method for predicting user-customized disease, comprising at least one of a recall index and an F1 score index. ◈Claim 11 was abandoned when paying the registration fee.◈ 11. The method of claim 10,
The user input information is
including the user's personal data, medical treatment data, and lifestyle data;
The user's personal data includes at least one of name, age, date of birth, and gender;
The user's medical treatment data includes at least one or more of medical treatment date, examination type, past medical history data, recent medical history data, recent department data, family history, and drug treatment history data;
The user's lifestyle data is a user-customized disease prediction method comprising at least one of drinking data, smoking data, hepatitis B antigen holding data, and current drug intake data. ◈Claim 12 was abandoned when paying the registration fee.◈ 12. The method of claim 11,
The disease occurrence data of the insurers includes cancer occurrence data, heart disease occurrence data, brain disease occurrence data, liver disease occurrence data, diabetes occurrence data, hypertension occurrence data, thyroid disease occurrence data, renal gland occurrence data, glaucoma occurrence data, and arterial disease occurrence data. curing occurrence data;
The output variable is a user-customized disease prediction method that is set to include at least one or more data among the data included in the disease occurrence data. ◈Claim 13 was abandoned when paying the registration fee.◈ 13. The method of claim 12,
The user disease prediction information is
Dangerous disease prediction information related to at least one or more of the data included in the disease occurrence data of the insurers, self-diagnosis method information corresponding to the dangerous disease prediction information, and prevention method information corresponding to the dangerous disease prediction information. User-tailored disease prediction methods. ◈Claim 14 was abandoned when paying the registration fee.◈ 14. The method of claim 13,
Step iii) is,
When the user disease prediction information includes a plurality of dangerous disease prediction information related to two or more data among the data included in the disease occurrence data, the server classifies the plurality of risk disease prediction information into a high-risk group and a low-risk group A user-customized disease prediction method comprising the step of providing to the terminal. ◈Claim 15 was abandoned when paying the registration fee.◈ 12. The method of claim 11,
After step iii) above,
The method further comprising the step of providing, by the server, insurance product recommendation information corresponding to the user's disease prediction information to the terminal. ◈Claim 16 has been abandoned at the time of payment of the registration fee.◈ A system for providing customized disease prediction information to a user through a communication connection with a terminal,
a communication module for transmitting and receiving information to and from the terminal;
a memory for storing a program for providing user-customized disease prediction information; and
A processor for executing the program stored in the memory;
The processor executes the user-customized disease prediction information providing program,
Receive user input information from the terminal, and based on a learning data set including personal data, medical treatment data, lifestyle data, and disease occurrence data of insurers received from the National Health Insurance Corporation server, the personal data, medical treatment data and Using the disease prediction model learned by using lifestyle data as an input variable and the disease occurrence data as an output variable, generating user disease prediction information based on the user input information, and converting the user disease prediction information to the terminal provide to,
The disease prediction model performs performance evaluation including accuracy evaluation on each of a plurality of artificial intelligence models based on test data including some of the data included in the training data set, and among the plurality of artificial intelligence models Selecting an artificial intelligence model having the highest performance index, and learning the selected artificial intelligence model based on the input variable and the output variable,
The performance evaluation includes at least one of an accuracy evaluation, a precision evaluation, a recall evaluation, and an F1 score evaluation for each of the plurality of artificial intelligence models, and the performance indicator includes an accuracy indicator corresponding to the performance evaluation, a precision indicator, A user-customized disease prediction system comprising at least one of a recall index and an F1 score index. ◈Claim 17 was abandoned when paying the registration fee.◈ 17. The method of claim 16,
The user input information is
including the user's personal data, medical treatment data, and lifestyle data;
The user's personal data includes at least one of name, age, date of birth, and gender;
The user's medical treatment data includes at least one or more of medical treatment date, examination type, past medical history data, recent medical history data, recent department data, family history, and drug treatment history data;
The user's lifestyle data is a user-customized disease prediction system including at least one of drinking data, smoking data, hepatitis B antigen holding data, and current drug intake data. ◈Claim 18 was abandoned when paying the registration fee.◈ 18. The method of claim 17,
The disease occurrence data of the insurers includes cancer occurrence data, heart disease occurrence data, brain disease occurrence data, liver disease occurrence data, diabetes occurrence data, hypertension occurrence data, thyroid disease occurrence data, renal gland occurrence data, glaucoma occurrence data, and arterial disease occurrence data. curing occurrence data;
The output variable is set to include at least one or more data among data included in the disease occurrence data, and
The user disease prediction information includes risk disease prediction information related to at least one data among data included in the disease occurrence data of the insurers, self-diagnosis method information corresponding to the risk disease prediction information, and a prevention method corresponding to the dangerous disease prediction information A user-customized disease prediction system that includes information. ◈Claim 19 was abandoned at the time of payment of the registration fee.◈ 19. The method of claim 18,
The processor executes the user-customized disease prediction information providing program,
When the user disease prediction information includes a plurality of dangerous disease prediction information related to two or more data among the data included in the disease occurrence data, a plurality of dangerous disease prediction information is classified into a high-risk group and a low-risk group and provided to the terminal wherein the user-customized disease prediction system is further configured to: ◈Claim 20 was abandoned when paying the registration fee.◈ 17. The method of claim 16,
The processor executes the user-customized disease prediction information providing program,
The user-customized disease prediction system that is further configured to provide the terminal with insurance product recommendation information corresponding to the user disease prediction information.

Images