KR20180061553A - System and method for predicting future health trends through prediction models based on similar case clustering - Google Patents

Abstract

The present invention relates to a system for predicting a future health trend through a predication model based on a similar case cluster and a method thereof, and more particularly, to a system for predicting a future health trend and a method thereof, capable of extracting a multiple association feature similar case cluster matched to a prediction query for the health information of a user through a future value prediction model and a class prediction model for a health characteristic of the similar case cluster generated by circularly clustering a target feature that is a health characteristic for personal health information and association features of the target feature, predicting the future health trend for each association feature using a multi-prediction model based on the corresponding similar case cluster, and outputting a prediction result by performing the ensembling of the prediction result. Accordingly, the present invention can simplify the configuration of the system for predicting a future health trend and quickly and reliably provide the prediction result to the user.

Description

TECHNICAL FIELD The present invention relates to a system and a method for predicting future health trends using a cluster-based prediction model,

The present invention relates to a system and method for predicting future health trends through a similar case cluster-based prediction model, and more particularly to a system and method for predicting future health trends through a similar case cluster- Based on the similar case cluster, the multi-associative feature similar case cluster matching the prediction query for the user's health information is extracted through the class prediction model and the future value prediction model of the health feature of the created similar case cluster A future health trend prediction system for predicting future health trends by association feature using a multiple prediction model, and outputting the predicted results by ensemble, and a method thereof.

With the recent development of medical science and the improvement of living standards, human life span is rapidly increasing, and modern society is turning into an aging society. On the other hand, new and diverse forms of disease are emerging due to global warming, increased risk factors for human health, and changes in lifestyle including eating habits.

Unlike the past, in which infectious diseases mainly occurred, the incidence of non-infectious diseases such as circulatory diseases, diabetes, cancer, cardiovascular diseases and hypertension increased rapidly have. These non - infectious diseases are mostly diseases with high treatment cost, so it is necessary to prevent and manage the health deterioration by predicting the future health trend of users. Therefore, the importance of preventing and managing prevention of health deterioration by predicting future health trends has been greatly emphasized.

As a result, there is a growing awareness of public health and the demand for medical services for proactive treatment in post-treatment of diseases is increasing rapidly. To solve this problem, Future health trend prediction systems that manage disease in advance are being developed. In order to efficiently manage the user's diseases through the future health trend prediction system, it is very important to accurately predict the future trend of the user's future health.

In general, the future health trend prediction system searches personal health records of people who show health status similar to the user's current health status based on the user's personal health record (PHR), which is time series data, Predict the future health status of the user by referring to changes in the health status of the persons concerned.

However, the conventional future health trend prediction system searches for similar cases based on all the health features (i.e., features) appearing in the user's personal health record, so that there are too many cases, Because the time is very long, the complexity of the system configuration is very high, and the search for similar cases also includes features with low relevance to the user's disease, the result of predicting the health status of the user based on the retrieved similar cases, The reliability of prediction can not be provided.

For example, even if only 10 cases are covered for each of the 37 features, a total of 10 ³⁷ cases should be dealt with. If the future health trends are to be handled for time series health record data, the time required for predicting health trends And the cost is high, and the accuracy of prediction is low.

Thus, in the present invention, clustering for a target feature (e.g., blood sugar) to predict a user's future health value and clustering for an association feature (e.g., systolic blood pressure, cholesterol, family history, etc.) The present invention provides a device and a method for providing reliable prediction results by rapidly reducing the complexity of a system while rapidly estimating future health.

That is, a class prediction model and a future value prediction model for health characteristics of the similar case cluster are generated using a similar case cluster generated by cyclically clustering a target feature, which is a health characteristic of health information, A plurality of class prediction models having high accuracy are selected from among the generated class prediction models and stored together with the future value prediction model (multiple prediction models) A system for extracting prediction results for a prediction prediction query, ensuring the extracted prediction results, extracting final class prediction probabilities, and predicting future health trends for the queries using a future value prediction model for the classes; I want to provide that method.

In addition, by using a hierarchical clustering technique, a plurality of similar case clusters are created, a prediction model is generated based on the generated similar case clusters, and the future health trend of the user is accurately predicted using the generated prediction model, The present invention provides an apparatus and method for efficiently performing prediction and management.

Next, a brief description will be given of the prior arts that exist in the technical field of the present invention, and technical matters which the present invention intends to differentiate from the prior arts will be described.

Korean Patent Laid-Open Publication No. 2016-0062668 (June 26, 2016) discloses an open health management apparatus and method, which searches for similar cases of user's health data by mining big data for each health case, The present invention relates to an open health management apparatus and method for predicting a future health state of a user based on similar cases and suggesting guidelines for health improvement.

The prior art has some similarities with the technical features of the present invention in that it searches for similar cases based on the user's health data and predicts the future health state of the user based on the similarity cases, Stage clustering for a particular target feature to predict a health value for a user rather than performing a two-step clustering on associated features associated with the target feature based on the performed one-step clustering At least one predictive model having a high prediction accuracy is selected from among the predictive models for each of the related features generated by performing the predictive modeling and predictive values for the specific target features predicted by the selected predictive model are ensured and predicted, The technical features of the present invention are not described or suggested .

Korean Patent Laid-Open Publication No. 2014-0022641 (Feb. 25, 2014) discloses an intelligent agent-based health log service system and method for managing chronic diseases. The system includes a biometric information database, personal information, The present invention relates to a system and a method for collecting information on an individual by using fuzzy inference and providing the collected information to the individual.

The prior art is somewhat similar to the present invention in that it provides information about an individual's health status based on individual health data. On the other hand, the present invention specifies a target feature for predicting a future health condition based on a user's personal health record, and hierarchically performs clustering for a plurality of related features associated with the specific target feature and the target feature And generating an optimal multiple prediction model for the target feature on the basis of the prediction result, thereby predicting a transition to future health of a user with a low complexity and quick and reliable user through the system. The above technical features are not described or suggested.

Disclosure of Invention Technical Problem [8] Accordingly, the present invention has been made in view of the above problems, and it is an object of the present invention to provide a health information prediction method and a health information prediction method for predicting similar cases using a prediction model based on similar case clusters, Health trend prediction system and method thereof.

The present invention also provides a method for classifying a similar case cluster for a prediction query for health information of a user, searching for a class prediction model for the identified similar case cluster, performing similar case prediction for each of a plurality of class prediction models, And outputting a class prediction result, thereby providing a system and a method for predicting future health trends that are excellent in processing speed and accuracy.

The present invention also provides a method for predicting future predictions of at least one future value prediction model by performing an ensemble on a plurality of class prediction results to select and output at least one future value prediction model, Health trend prediction system and method thereof.

The present invention also includes a method of generating a plurality of similar case clusters for a target feature using a hierarchical clustering technique when performing similar case clustering to predict future health trends for a specific target feature, And a similar case clustering method for generating similar case clusters for association features associated with the related features, and a method for predicting future health trends that can reduce the complexity of the configuration drastically.

In addition, the present invention selects a best-fit association feature for the target feature based on the similar-case cluster generated through the hierarchical clustering, thereby rapidly retrieving the similar-case cluster for the user's target feature and the similar- And to provide a system and a method for searching for a future health trend of a user.

The present invention also relates to a method for estimating a future state of health of a user, comprising the steps of: clustering a target feature to predict a future health state of a user; and clustering the related features associated with the target feature based on the clustering, An object of the present invention is to provide a prediction result of a reliable user's future health by selecting a prediction model and ensuring at least one prediction result outputted through the selected prediction model to obtain a final prediction result .

A future health trend prediction system using a similar case cluster based prediction model according to an embodiment of the present invention includes a class prediction model selection unit for selecting a plurality of class prediction models from a prediction query for health information of a user, A class and a future value predicting unit for performing a model-by-model prediction to output a plurality of class prediction results, performing prediction for at least one or more future value prediction models to output a future value prediction result, And a future value prediction model selector for selecting and outputting at least one or more future value prediction models by performing an ensemble.

Here, the class prediction model selection unit may include a similar case cluster discrimination unit for discriminating the similar case cluster by receiving a prediction query for the user's health information, and a class prediction model searching unit for searching the class prediction model for the determined similar case cluster Wherein the similar case cluster is generated by hierarchically clustering a target feature and an association feature of the target feature from a plurality of time series health data and grouping a plurality of patterns that are generated in a time- And a target feature appearing after a predetermined time period included in the similar case cluster are classified into a plurality of classes.

Wherein the class prediction model is a prediction model for a probability of a class in a similar case cluster for the association feature, the future value prediction model further comprising: Model, or all the classes of the similar case cluster for the association feature.

Here, the prediction of the future health trend is characterized by predicting a change in the future health trend of a section following a change pattern for a specific section of time-series health data.

In addition, the similar-case cluster discrimination unit judges the similar case cluster by matching the representative information on the similar case cluster with the prediction query, and the representative information forms one similar case cluster and one similar case cluster And is information on a change pattern representing a plurality of time series personal health data included. In addition, the similar case cluster discrimination unit judges the similar case cluster by matching the health feature of the related feature cluster selected from the class prediction models of the similar case cluster with the prediction query, And an associated feature extracted in the process of selecting an associated feature class prediction module that satisfies a criterion for a predetermined accuracy among the features.

The class prediction model searching unit may further include searching and loading a prediction model for a similar case cluster determined to match the prediction query from the class prediction model database.

According to another aspect of the present invention, there is provided a method for predicting future health trends through a similar case cluster-based prediction model, the method comprising: a class prediction model selection step of selecting a plurality of class prediction models from a prediction query for health information of a user; A class and a future value prediction step of outputting a plurality of class prediction results by performing prediction by class prediction model, outputting a future value prediction result by performing prediction of at least one or more future value prediction models, And selecting and outputting at least one or more future value prediction models by performing an ensemble on the future value prediction model.

In another aspect of the present invention, there is provided a method for predicting future health trends through a similar case cluster-based prediction model, the method comprising: determining whether a corresponding prediction model of an associated feature cluster matching a prediction query for health information of a user satisfies a predetermined accuracy A prediction class prediction step of classifying the plurality of prediction models and outputting a prediction result of each prediction model and a prediction value prediction step of classifying the plurality of prediction results, And outputting a prediction result by performing an ensemble.

As described above, the future health trend prediction system and the method for predicting future health trends through the similar case cluster-based prediction model of the present invention can be realized by cyclically clustering the target feature and the association feature of the target feature based on a plurality of personal health record data A class prediction model and a future value prediction model for the health characteristics of the similar case cluster are generated using a similar case cluster, and a plurality of class prediction models with high accuracy are selected from the generated class prediction models, Extracting a prediction result for a specific future health transition prediction query using the class prediction model among the multiple prediction models from a prediction query for the health information of the user, Ensemble the result, extract the final class prediction probability, The future health trend prediction for the query is predicted using the future value prediction model for the query, and the structure of the future health trend prediction system can be simplified, and quick and reliable prediction results can be provided to the user.

FIG. 1 is a conceptual diagram for schematically explaining a future health transition prediction system and a method thereof according to a similar case cluster-based prediction model according to an embodiment of the present invention.
2 is a diagram schematically illustrating a process of generating a prediction model by performing hierarchical similar case clustering according to an embodiment of the present invention.
3 is a block diagram illustrating a configuration of a future health transition prediction apparatus according to an embodiment of the present invention.
FIG. 4 is a view for explaining individual time-series personal health record data according to an embodiment of the present invention.
5 is a flowchart illustrating a similar case clustering process of generating information utilized in an embodiment of the present invention and a process of generating a future health trend prediction model.
FIG. 6 is a flowchart illustrating a procedure of a future health transition prediction process according to an embodiment of the present invention.
7 is a flowchart illustrating a procedure for predicting a future health trend according to another embodiment of the present invention.

Hereinafter, a system and method for predicting future health trends through a similar case cluster-based prediction model of the present invention will be described in detail with reference to the accompanying drawings. The present invention may be embodied in many different forms and is not limited to the embodiments described herein. Like parts are designated with like reference numerals throughout the specification.

Hereinafter, a system and method for predicting future health trends through a similar case cluster-based prediction model will be briefly described.

FIG. 1 is a conceptual diagram for schematically explaining a future health transition prediction system and a method thereof according to a similar case cluster-based prediction model according to an embodiment of the present invention.

1, the future health transition prediction system 10 includes a similar case clustering apparatus 100 for performing similar case clustering, a future health transition prediction model generating apparatus 200 for generating a future health transition prediction model, And a future health transition prediction apparatus 300 that predicts and provides a future health transition for a target feature of a user according to a user's query using the generated prediction model.

The future health trend prediction system 10 further includes a similar case cluster database 500a, a cluster representative information database 500b, a predictive model database 600a, and a target optimal relevance feature database 600b on a network or locally .

Here, the similar-case clustering apparatus 100 receives personal time-series health information from the personal health record database 400, and performs clustering on similar cases. The result of the clustering is stored in the similar case cluster database 500a, and the set of representative information of the similar case cluster is stored in the cluster representative information database 500b.

The similar case cluster is input to the future health trend prediction model generating apparatus 200 and used for generating a prediction model for predicting future health trends, the generated prediction model is stored in a prediction model database 600a, The prediction model is used to predict future health trends in the future health trend prediction apparatus 300 together with the representative information of the similar case cluster. This process can be controlled by the user and the administrator through the communication network, and the prediction result is also provided to the user.

Hereinafter, a process of generating a similar case cluster by performing similar case clustering using the target feature as the blood sugar, and generating a prediction model therefor will be described.

2 is a diagram schematically illustrating a process of generating a prediction model by performing hierarchical similar case clustering according to an embodiment of the present invention.

The similar case clustering apparatus 100 performs hierarchical clustering of a plurality of personal health records loaded from the personal health record database 400 to generate a similar case cluster to generate a similar case prediction model.

Wherein the hierarchical clustering includes performing similar case clustering for the target feature and performing similar case clustering for the association feature having a close association with the target feature. On the other hand, the similar case clustering can be performed for all health features that appear in a personal health record, as well as being performed for an association feature that is closely related to the target feature.

Also, the target feature may refer to a health feature (e.g., blood glucose) that is subject to future health trend predictions among the health features included in a plurality of personal health records, and the associated feature may include a health feature associated with the target feature it means. For example, if the target feature is blood sugar, the associated features may be systolic blood pressure, diastolic blood pressure, LDL cholesterol, family history (e.g., diabetes), and the like.

The personal health record database 400 may be implemented on a local or network and may be public cohort information provided by the HIRA or the NHS or may be a patient's personal health record provided by a medical institution such as a hospital, It is a repository for storing individual personal health records, such as provided personal health records. The personal health record may also be grouped according to gender, age, etc. and stored in the personal health record database 400.

More specifically, the similar-case clustering apparatus 100 has a plurality of personal health records (e.g., a plurality of personal health records) loaded from a personal health record database 400 so as to efficiently perform learning for generating similar- Into a time series of personal health record data and normalize it. The converted individual health record data is obtained by grouping and converting the respective health characteristics that are changed in a time series on the basis of each personal health record, and is obtained by converting the health characteristics (for example, weight, Height, systolic blood pressure, diastolic blood pressure, age, etc.).

Also, the similar-case clustering apparatus 100 performs similar-case clustering on the target feature based on the individual health record data (step 1) to generate a prediction model for predicting the future health trend for the target feature, A plurality of similar case clusters (e.g., 10 clusters) are created. Clustering is performed on the related features for each of the generated one-stage similar case clusters to generate a plurality of similar case clusters.

Clustering is performed on the associated features associated with the target feature based on the individual health record data included in the similar case clusters for each of the generated target features to obtain a plurality of associated features for each similar feature cluster for each target feature Create a similar case cluster.

That is, the similar case clustering for the target feature is to group the target features, which are time-sequentially changed for a predetermined time period, by pattern, to generate a plurality of similar case clusters for the target feature, The clustering is to generate a plurality of similar case clusters for each of the associated features associated with the target feature for each similar case cluster for the target feature with the same mechanism as the similar case clustering for the target feature.

In the present invention, the hierarchical clustering for creating a similar case cluster is not limited to performing the one-stage and two-stage similar case clustering, and it is also possible to perform clustering for one- A similar case cluster may be created by carrying out two-step clustering and performing similar-case clustering in a plurality of steps, such as performing three-step clustering with another related feature associated with the association feature used for two-step clustering .

Therefore, the similar-case clustering apparatus 100 creates a similar case cluster by performing hierarchical clustering on a plurality of personal health records stored in the individual health record database 400, stores the similar case cluster in the similar case cluster database 500a, The representative information on the similar case cluster is stored in the similar case cluster representative information database 500b. In other words, clustering is performed by extracting a time-series change pattern of a specific section for the target feature from the time-series health data, and then, for the related feature for each cluster of the target feature, The clustering is performed by extracting a change pattern of a specific period, and predicting the future health trend is to predict a pattern of a future health trend of a section following the time varying pattern of the specific period. Wherein the change pattern is representative information on a cluster of the target feature and the association feature, and the representative information is used to identify a similar case cluster for a health characteristic of the user when a query for future health transition prediction is input from the user , And forms a similar case cluster, and is information of a change pattern representing a plurality of personal health data included in one similar case cluster.

Also, the future health transition prediction model generation apparatus 200 learns each of the similar case clusters for each of the related features generated through the hierarchical clustering technique, and generates a cluster-specific prediction model for the target feature. Selecting a prediction model having an accuracy higher than or equal to a preset numerical value or having an accuracy higher than or equal to a predetermined rank by testing each of the generated prediction models to thereby select an optimal related feature for a specific target feature, And stores a prediction model in the prediction model database 500a.

In addition, the future health trend prediction model generating apparatus 200 learns a plurality of similar case clusters clustered in a hierarchical clustering technique from time series personal health record data through machine learning to generate a prediction model. Wherein the hierarchical clustering technique creates a similar case cluster for a target feature from personal health record data and creates a similar case cluster for an associated feature associated with the target feature, And a target feature appearing after a predetermined time period included in the similar case cluster are classified into a plurality of classes. Generating a class prediction model by predicting a probability for each class based on a linear or nonlinear distribution of individual health record data included in each of the similar case clusters for the associated feature, A future value prediction model for predicting a future value for the feature is generated and stored in the prediction model database 600a. Then, the generated class prediction model is tested to select a class prediction model having an accuracy higher than or equal to a preset numerical value or having an accuracy higher than a predetermined order, and stored in the prediction model database 600a.

Information about a set of the optimal association features for each target is extracted for the selected class prediction model and stored in the target-specific optimum association feature database 600b. In addition, the future health trend prediction apparatus 300 provides information on an optimal association feature so that a prediction model of a cluster matching a prediction query for a user's specific health information can be selected. Herein, the accuracy is calculated by calculating information on a class prediction success or a class prediction probability from the class prediction result using predetermined test input data not used to generate the class prediction model, It is cumulatively repeatedly calculated with respect to the input data.

Accordingly, the present invention can be applied not only to search for similar cases for all health characteristics appearing in the individual health record, but also to search for similar cases preferentially for specific health characteristics to be predicted for future health trends, Since similar cases are searched only for optimal health characteristics that are related to specific health characteristics among similar cases of specific health characteristics, the number of cases for the search can be drastically reduced, and the future health trend prediction system The complexity of the configuration of the display device 10 can be drastically reduced.

It also predicts the future health trends for the target features by using the best related features that are closely related to the target features, rather than predicting future health trends based on all the health characteristics shown in the individual health record. It is possible to shorten the time required for the prediction, and to provide reliable prediction results with high accuracy.

Hereinafter, the future health transition prediction apparatus 300 will be described in detail with reference to FIG. 3 is a block diagram illustrating a configuration of a future health transition prediction apparatus according to an embodiment of the present invention.

3, the future health trend prediction apparatus 300 includes a class prediction model selection unit 310 for selecting a plurality of class prediction models from a prediction query for the user's health information, a plurality of class prediction models A class and future value predicting unit 320 for performing similar case prediction to output a plurality of class prediction results, performing similar case prediction for at least one or more future value prediction models to output a future value prediction result, And a future value prediction model selector 330 for performing an ensemble on the class prediction result to select and output at least one or more future value prediction models.

The class prediction model selection unit 310 includes a similar case cluster discrimination unit 311 for receiving a prediction query for health information of a user to discriminate a similar case cluster and a class prediction model for the determined similar case cluster Wherein the similar case clusters are generated by hierarchically clustering a target feature and an association feature of the target feature from a plurality of time series health data, And a target feature appearing after a predetermined time period included in the similar case cluster are classified into a plurality of classes.

In addition, the future health trend prediction apparatus 300 receives a user query and a personal health record from a user through a user interface (not shown), performs a preprocessing process, and generates and normalizes time series personal health record data. This is performed in the same manner as the preprocessing process performed in the similar-case clustering apparatus 100.

That is, the preprocessing process loads a plurality of personal health records stored in the personal health record database 400 and converts them into individual time-series personal health record data. The converted individual health record data is converted to efficiently generate the similar case cluster and the learning of the prediction model, and is configured to include each health characteristic according to a predetermined time period. The individual time-series personal health record data described above is shown in FIG. And normalizes the converted individual health record data so as to have a value between 0 and 1. In addition, health characteristics that do not appear as specific numerical values such as smoking, drinking or not, can be normalized to 0 or 1.

As shown in FIG. 4, a personal health record of a particular individual. A plurality of personal health records may be stored in the personal health record database 400 as unique values indicating respective health information, as shown in (a) of FIG. 4 , And it is loaded in the preprocessing unit according to the present invention and converted into individual time-series personal health record data as shown in FIG. 4 (b).

Also, the preprocessing process selects data available for prediction model learning from the converted individual health record data. It is selected to learn health characteristics that change over a period of time and to predict trends in health characteristics after the predetermined period of time. For example, the learning data used to predict changes in blood glucose levels over a three-year period to predict blood glucose levels at 4 years requires a four-year continuous measurement of blood glucose levels, That is, four consecutive years of screening). It is possible to check whether the data is missing from the selected personal health record data. As a result of the check, if there is missing data, the missing data can be interpolated by calculating the median value or the average value. For example, if blood glucose levels in years 13, 15, and 16 are 80 mg / dl, 90 mg / dl, and 95 mg / dl, respectively, and the blood glucose levels in year 14 are missing, Based on the blood glucose level of the year, the median or average value can be calculated to interpolate the blood glucose level at 15 years to 85 mg / dl.

The similar case cluster discrimination unit 311 loads a similar case cluster for a specific target feature according to a query of a user and a similar case cluster for each related feature for each similar case cluster for the target feature. On the other hand, the similar case cluster for each related feature for each similar case cluster to the target feature is a similar case cluster for the best related feature for a specific target feature selected by the future health transition prediction model generating apparatus 200.

The similar case cluster discrimination unit 311 analyzes a pattern of the target feature indicated in the personal health record data of the preprocessed user to determine a first-stage similar case cluster for the loaded target feature. And analyzes the pattern for each related feature indicated in the personal health record data of the preprocessed user to determine the similar case cluster for each loaded feature. On the other hand, it is a matter of course that the personal health record data included in the identified similar case cluster per related feature is a set of individual health record data included in the selected similar case cluster.

The class prediction model searching unit 312 searches for and loads the prediction model database 500a storing the class prediction model generated by learning the similar case cluster for each related feature determined as described above. The loaded class prediction model is an optimal multiple prediction model selected by the future health trend prediction model generating apparatus 200. [

The class and future value predicting unit 320 includes a class predicting unit 321 for performing similar case prediction for each of a plurality of class prediction models to output a plurality of class prediction results and a similar case prediction And a future value predicting unit 322 for outputting a future value prediction result.

In particular, the class predicting unit 321 predicts the probability of each class for the target feature through the loaded class prediction model, and outputs the prediction result for each class. Wherein the class prediction model is a prediction model for a probability of a class in a similar case cluster for the association feature, the future value prediction model further comprising: a future value prediction model learning for each class of the similar case cluster for the association feature Or all the classes of the similar case cluster for the related feature.

In addition, predicting the future health trend is to predict a change in the future health trend of a section following a change pattern for a specific section of time series health data.

The future value prediction model selection unit 330 includes a prediction result ensemble unit 331 and a future value prediction model search unit 332.

Next, referring to FIGS. 5 to 7, a description will be made of a similar case clustering, a future health trend prediction model generation process, and a future health trend prediction process.

Clustering of similar cases and prediction of future health trends We will explain how information is generated and used because it predicts future health trends by using information and prediction models generated through model generation process. Since the utilization process of the information may be changed according to the process of generating the information, the process of generating the information will be described first.

5 is a flowchart illustrating a similar case clustering process of generating information utilized in an embodiment of the present invention and a process of generating a future health trend prediction model.

As shown in FIG. 5, the similar case clustering apparatus 100 receives the personal health record from the personal health database 400 and performs preprocessing. The pre-processing includes converting the input personal health record data into a format suitable for use in the similar case clustering apparatus 100, and performing a process of normalizing each feature. The conversion may include converting various information such as text, numbers, images, or voice to be suitable for clustering according to the type of data. The normalization converges the dynamic range of input data to a value between 0 and 1, and simplifies and unifies the handling of data through the normalization.

The preprocessed personal health record also creates a similar case cluster by hierarchical clustering. Since the generated similar case clusters have various hierarchical structures, they are stored in the similar case cluster database 500b while maintaining the characteristics of the hierarchical structure. For example, if the target feature is blood sugar, grouping a plurality of target features having a pattern of similar pattern, based on the pattern for the time-varying blood glucose levels of the blood glucose measured over three years, (Pattern) information for the target feature of the target. Such a representative pattern is used to identify a similar case cluster for a health characteristic of a user when a query for prediction of future health transition is input from a user and forms one similar case cluster, And represent a plurality of personal health data included in the case cluster.

Here, the similar case cluster generation process by hierarchical clustering first performs the target feature clustering and generates representative information for each target feature, and when the query for future health transition prediction is input from the user, A similar case cluster is used to identify a similar case cluster for a health characteristic of the user and is a pattern information representing a plurality of personal health data included in one similar case cluster. Next, an association feature clustering for the individual health records contained in the cluster for each target feature is performed, and the cluster representative information is also calculated for the performed association feature cluster. Next, each cluster generated through the association feature clustering is classified into classes and a probability distribution of personal health record data for the class is calculated.

The generated similar case cluster representative information is stored in the similar case cluster representative information database 500b, and the similar case cluster is stored in the similar case cluster database 500a.

Next, in the process of generating the future health trend prediction model, the similar case cluster stored in the similar case cluster database 500a is first loaded by loading it in the future health trend prediction model generating apparatus 200. [ Next, the similar case cluster-specific prediction model learning is performed to generate a prediction model for each of the similar case clusters. Similar case cluster-by-cluster prediction model learning is performed for a plurality of similar case clusters generated for each related feature for a similar case cluster of a specific target feature. In this process, in a similar case cluster for a target feature and a similar case cluster for an association feature associated with the target feature, a similar model cluster for the related feature is learned to generate a prediction model.

For example, if the target feature is blood glucose, a similar case cluster for blood glucose is generated, and if the associated features for blood glucose are systolic blood pressure, diastolic blood pressure, and cholesterol, then the systolic blood pressure, diastolic blood pressure , A similar case cluster for cholesterol is generated, and then a plurality of similar case clusters generated for each of the systolic blood pressure diastolic blood pressure cholesterol are learned to generate a prediction model for the blood glucose.

Similar case clusters for associative features are categorized by class. Here, the class prediction model is divided into a class prediction model for predicting the class probability in the cluster, and a future value prediction model for predicting a future value for each class of each cluster or for the entire cluster.

Therefore, the future value prediction model is generated for each similar case cluster or each class in the similar case cluster and stored in the similar case prediction model database 600a.

However, instead of storing all the class prediction models in the similar case prediction model database 600a, the optimum prediction models are selected and stored among the plurality of class prediction models. The selection may be performed by testing a plurality of prediction models for each associated feature to calculate an accuracy for each prediction model and selecting a plurality of prediction models having an accuracy higher than a preset value or having an accuracy higher than a predetermined order, .

On the other hand, the test is performed on all the class prediction models generated by learning clusters for each related feature, and the accuracy for all the class prediction models is calculated using data used for learning. This allows the selection of the most accurate class prediction models to select the best-fit associative features that are closely related to future changes in a particular target feature.

For example, if the target feature is blood sugar and the associated features to blood glucose are systolic blood pressure, LDL cholesterol, the future health prediction model generation device 200 determines similar case clusters for blood glucose of the test data, A cluster of similar cases for all relevant features (i.e., systolic blood pressure, LDL cholesterol) is determined. Thereafter, the predicted model of systolic blood pressure and LDL cholesterol is loaded from the predictive model database 600a, the predicted results of the class predictive model are compared with the actual values of the test data by inputting the plurality of test data, And calculates the accuracy of the loaded class prediction model.

That is, it is possible to judge the success or failure of prediction by comparing the class prediction result with the class of the inputted test data, or to select the upper several classes having a high probability of prediction. By performing such a mechanism for all the test groups, And the accuracy of the class prediction model for LDL cholesterol. This is performed for a candidate probability prediction model generated for each similar case cluster for each related feature, thereby selecting an optimal class prediction model.

In addition, since the class prediction model is generated for each related feature, the determination of the class prediction model with high accuracy through the above-described series of processes is performed by selecting a related feature closely related to a future change of a specific target feature It has the same effect.

Next, we describe the process of predicting future health trends by using representative information of similar case clusters generated in this way, aggregation of optimal relevance features for each target, and similar case prediction models.

FIG. 6 is a flowchart illustrating a procedure of a future health transition prediction process according to an embodiment of the present invention.

As shown in FIG. 6, the procedure for predicting future health trends by the future health trend predicting apparatus 300 is as follows. First, when a user inputs a future health trend prediction query for a specific target feature together with a personal health record S310), the future health trend prediction apparatus 300 matches the prediction query of the user based on the representative information (pattern information and the like) of the similar case cluster and the information on the optimum association feature set per target, (S320). That is, in order to efficiently perform the future health transition for the target feature, the future health trend predicting device 300 may similarly classify the input personal health record as the time series personal health record data , Normalizes the converted individual health record data, and identifies each similar case cluster. The determination may comprise analyzing a pattern for a predetermined period of time for a target feature and a plurality of associated features from the individual health record data to determine a similarity measure for a similar feature cluster among similar case clusters for a particular target feature that is pre- Determining a case cluster, and then determining a similar case cluster per association feature with the same mechanism. On the other hand, the association feature for the target feature means the optimal relevance feature selected in the future health trend prediction model generation apparatus 200.

Next, the future health trend prediction apparatus 300 searches the similar case prediction model database 600a using the query matching cluster index according to the similar case cluster determined to match the prediction query, The multi-class prediction model is retrieved and loaded (S330).

The multi-class prediction model for each similar case cluster for the associated features loaded through the search is an optimal class prediction model selected through accuracy calculation. For example, if the target feature is blood sugar and the optimal associated feature for blood glucose selected by the future health trend prediction model generating device 200 is systolic blood pressure, diastolic blood pressure, LDL cholesterol, the future health trend prediction device 300 ) Identifies a similar case cluster for blood glucose from the user's personal health record data, and the discrimination for the similar case cluster for the related feature of the identified similar case cluster is performed only for systolic blood pressure, diastolic blood pressure, and LDL cholesterol do. Thereafter, the optimal class prediction model generated by learning the similar systolic blood pressure, diastolic blood pressure, and LDL cholesterol similarity clusters is loaded. That is, at least one or more class prediction models can be determined when determining the optimal class prediction model by calculating the accuracy for the class prediction model, so that the class prediction model for at least one or more optimal association features for a particular target is selected .

Next, the future health trend prediction apparatus 300 performs class prediction for each model using the retrieved and loaded multi-class prediction model (S340). The result of the class prediction for each model is a prediction result on the probability for each class.

Next, the future health transition prediction apparatus 200 asserts the predicted class prediction results for each model, finally extracts a class prediction probability for the target feature of the user, and outputs an index of the final class at step S350.

Next, the future health trend prediction apparatus 300 loads the final value prediction model of the final class using the index of the final class (S360). Subsequently, the future value prediction model is extracted from the similar case prediction model database 600a to perform future value prediction, and the future value of the final class is output to the user (S370).

Here, the future health trend predicting apparatus 300 can predict the final health trend by averaging or calculating the median based on the predicted results of each model. That is, the future health trend predicting apparatus 300 determines a final prediction class by ensuring a plurality of class prediction results that predict a probability value for each class of the optimal association feature through a class prediction model, By using the model to predict the future value for the class, the final prediction result for the user's query can be output. On the other hand, as described above, the final prediction class can be determined to be at least one or more.

In addition, even when a plurality of optimal association features are selected, a class prediction model for the selected plurality of optimal association features is loaded, a final prediction class is determined for each of the loaded class prediction models, and the determined plurality of final prediction classes And predicts future values for the class using the future value prediction model for the class. Thereafter, the future health trend prediction apparatus 300 may provide the predicted plurality of future values to the user, or may provide the user with the predicted plurality of future values as an average or an intermediate value.

Hereinafter, a future health transition prediction process according to another embodiment of the present invention will be described.

7 is a flowchart illustrating a procedure for predicting future health trend according to another embodiment of the present invention.

As shown in FIG. 7, the process of loading a future health trend prediction query for a specific target feature together with a personal health record from a user and loading a corresponding prediction model of a similar case cluster matched with the prediction query is the same as before Lt; / RTI > Of course, it is not necessary to match and load the class prediction model with respect to the user's prediction query. The predictive query of the user may be input and the predictive model may be loaded immediately without distinguishing the class predictive model or the future predictive model (S310 to S330).

A process of reducing the number of models to be predicted by filtering the loaded prediction models is required (S340a). The filtering may utilize the distribution of the target features and the associated features calculated in the similar case clustering process, the distribution of the classes, and the prediction probability values between the respective related features.

Subsequently, prediction is performed for each prediction model using the filtered prediction model (S350a). The filtered prediction model (a prediction model having the best probability value or a prediction model corresponding to a plurality of higher probability values) can be plural, and thus, a plurality of prediction results generated here are also obtained.

Next, the generated prediction results for each model are ensemble, and the prediction results are output to the user (S360a).

As described above, the future health trend prediction apparatus 300 extracts at least one prediction model through filtering for a plurality of similar case prediction models without class and future values, The final prediction result for the user query can be output by ensuring the extracted prediction results for the model and predicting the final future value.

As described above, the future health trend prediction system and the method using the similar case cluster-based prediction model of the present invention can perform hierarchical clustering based on a plurality of personal health record data, By generating a case cluster and predicting the future health trend of the user through the prediction model in which the generated similar case cluster is learned, it is possible to significantly reduce the complexity of the system configuration and provide a quick and reliable prediction result to the user There is an effect that can be.

It will be apparent to those skilled in the art that various modifications may be made to the invention without departing from the spirit and scope of the invention as defined in the following claims And changes may be made without departing from the spirit and scope of the invention.

10: Future health trend forecasting system
100: similar case clustering device 200: future health trend prediction model generating device
300: Future health trend prediction device 400: Personal health history database
500a: Similar case cluster database
500b: cluster representative information database
600a: Similar case prediction model database
600b: Optimal association feature database per target
310: Class prediction model selection unit 320: Class and future value prediction unit
330: future value prediction model selection unit 311: similar case cluster determination unit
312: class prediction model searching unit 321: class prediction unit
322: future value prediction unit 331: prediction result ensemble unit
332: prediction value prediction model search unit

Claims (15)

A class prediction model selection unit for selecting a plurality of class prediction models from a prediction query for health information of a user;
A class and a future value predicting unit for performing a plurality of class predictive model predictions to output a plurality of class predicted results, and to predict at least one or more future predictive models to output a future value predicted result; And
And a future value prediction model selector for performing an ensemble on the plurality of class prediction results to select and output at least one or more future value prediction models. The method according to claim 1,
Wherein the class prediction model selector comprises:
A similar case cluster discrimination unit for receiving the prediction query for the health information of the user and discriminating the similar case cluster; And
And a class prediction model retrieval unit for retrieving a class prediction model for the identified similar case cluster,
The similar case cluster is generated by hierarchically clustering a target feature and an association feature of the target feature from a plurality of time series health data, and generates a plurality of patterns that are generated by grouping a plurality of patterns that are changed in a time- Wherein the personal health record data and the target features appearing after a predetermined time period included in the similar case cluster are classified into a plurality of classes. 3. The method of claim 2,
Wherein the class prediction model is a prediction model for the probability of a class in a similar case cluster for the association feature,
The future value prediction model is a future value prediction model that is learned by including a future value prediction model learned for each class of the similar case cluster for the association feature or all classes of the similar case cluster for the association feature Future health trend prediction system based on cluster - based prediction model with similar case. The method according to claim 1,
To predict the future health trend,
A future health trend prediction system based on a similar case cluster based prediction model, characterized by predicting a change in future health trends of a section following a change pattern for a specific section of time series health data. 3. The method of claim 2,
The similar case cluster determination unit may determine,
Wherein the similar case cluster is determined by matching the representative information on the similar case cluster with the prediction query,
Wherein the representative information is information on a change pattern representing a plurality of time series personal health data included in one similar case cluster forming one similar case cluster, and a future health trend based on a similar case cluster based prediction model Prediction system. 6. The method of claim 5,
The similar case cluster determination unit may determine,
A similar case cluster is determined by matching the health feature of the selected feature feature cluster and the prediction query among the class prediction models of the similar case cluster,
Wherein the selected associated feature is an associated feature extracted in the process of selecting an associated feature class prediction model that satisfies a criterion for a predetermined accuracy among all of the associated features. Prediction system. The method according to claim 6,
Wherein the class prediction model searching unit comprises:
And retrieving and loading a prediction model for a similar case cluster identified as being matched with the prediction query from a class prediction model database, and loading the similar model cluster based prediction model. A class prediction model selection step of selecting a plurality of class prediction models from a prediction query for health information of a user;
A class and a future value prediction step of performing a plurality of class prediction model-by-model prediction to output a plurality of class prediction results, performing a prediction for at least one or more future value prediction models, and outputting a future value prediction result; And
And a future value prediction model selection step of performing an ensemble on the plurality of class prediction results to select and output at least one or more future value prediction models. . 9. The method of claim 8,
The class prediction model selection step includes:
A similar case cluster discrimination step of discriminating a similar case cluster by receiving a prediction query for health information of a user; And
And a class prediction model retrieval step of retrieving a class prediction model for the identified similar case cluster,
The similar case cluster is generated by hierarchically clustering a target feature and an association feature of the target feature from a plurality of time series health data, and generates a plurality of patterns that are generated by grouping a plurality of patterns that are changed in a time- Wherein the personal health record data and the target features appearing after a predetermined time period included in the similar case cluster are classified into a plurality of classes. 10. The method of claim 9,
Wherein the class prediction model is a prediction model for the probability of a class in a similar case cluster for the association feature,
The future value prediction model is a future value prediction model that is learned by including a future value prediction model learned for each class of the similar case cluster for the association feature or all classes of the similar case cluster for the association feature Future health trend prediction method using cluster - based prediction model with similar case. 9. The method of claim 8,
To predict the future health trend,
A method for predicting future health trends through a similar case cluster based prediction model, characterized by predicting a change in future health trends of a section following a change pattern for a particular section of time series health data. 10. The method of claim 9,
In the similar case cluster determination step,
Wherein the similar case cluster is determined by matching the representative information on the similar case cluster with the prediction query,
Wherein the representative information is information on a change pattern representing a plurality of time series personal health data included in one similar case cluster forming one similar case cluster, and a future health trend based on a similar case cluster based prediction model Prediction method. 13. The method of claim 12,
In the similar case cluster determination step,
A similar case cluster is determined by matching the health feature of the selected feature feature cluster and the prediction query among the class prediction models of the similar case cluster,
Wherein the selected associated feature is an associated feature extracted in the process of selecting an associated feature class prediction module that satisfies a criterion for a predetermined accuracy among all the related features. Prediction method. 13. The method of claim 12,
Wherein the class predictive model retrieving step comprises:
And retrieving and loading a prediction model for a similar case cluster identified as being matched with the prediction query from a class prediction model database and estimating future health trend based on the similar case cluster based prediction model. A predictive model filtering step of filtering the predictive model as to whether or not a corresponding predictive model of an associated feature cluster matching a predictive query for health information of a user satisfies a predetermined accuracy;
A class and a future value prediction step of performing a prediction on the plurality of filtered prediction models and outputting a prediction result of each prediction model; And
And outputting a prediction result by performing an ensemble on the plurality of prediction results. A method for predicting future health trends through a similar case cluster based prediction model.

Images