KR20200010621A - Device and method for insurance unfair claim and unfair pattern detection based on artificial intelligence - Google Patents

Abstract

According to an embodiment of the present invention, an apparatus for detecting a fraudulent insurance claim and a fraudulent pattern based on artificial intelligence includes: a data preprocessing part deriving property variables by standardizing insurance claim history data; an abnormal value detection part detecting a normal pattern and a fraudulent pattern from an abnormal group detected based on an abnormality detection algorithm into which the property variables are inputted; a new pattern classification part detecting a new pattern of a fraudulent claim by performing unsupervised learning into which the normal and fraudulent patterns are inputted; a new pattern detection part detecting a pattern of a new fraudulent claim by building a pattern determination model based on supervised learning into which the new pattern of the fraudulent claim and a pre-classified insurance claim pattern are inputted; and a fraudulent claim determination part determining whether a new insurance claim history is fraudulent by building a fraudulent claim detection model based on supervised learning into which the property variables and the normal claim data are inputted.

Description

Artificial intelligence-based insurance claim fraud and fraud pattern detection device and method {DEVICE AND METHOD FOR INSURANCE UNFAIR CLAIM AND UNFAIR PATTERN DETECTION BASED ON ARTIFICIAL INTELLIGENCE}

The present invention relates to an artificial intelligence-based claims and claims pattern detection apparatus and method.

The existing insurance fraud prevention system calculates rules based on business rules based on the reviewer's experience and knowledge, and divides them into investigation rules and exclusion rules. And as it is advanced, it is necessary to continuously update new fraudulent patterns or fraud patterns.

In addition, insurers cannot leave a claim without taking any action on insurance fraud, and cannot spend an infinite amount of investigation on the purpose of fully revealing the fraud. Thus, insurance companies need to reduce the amount of leakage from insurance fraud and determine the level of investigation efforts on a reasonable basis without spending too much investigative expenditure. Recently, technologies for analyzing insurance frauds have been developed through AI-based learning models, but the level of development is not appropriate for insurance fraud analysis techniques.

Background art of the present application is disclosed in Korean Patent Publication No. 10-0862181.

An object of the present invention is to solve the above-mentioned problems of the prior art, and to provide an artificial intelligence-based claim and claim pattern detection apparatus and method that provides a discrimination model that can determine a new pattern of claims unfair claim. do.

The present invention is to solve the above-mentioned problems of the prior art, artificial intelligence-based insurance claims unfair and unfair claims that can analyze the new unfair claims pattern by learning the unfair claims data, and determine the type of claim pattern for the new claim An object of the present invention is to provide a pattern detecting apparatus and method.

The present invention is to solve the above-mentioned problems of the prior art, and to provide an artificial intelligence-based claim unfair claims and fraud pattern detection apparatus and method for learning the claim-related data to determine unfairness for a new claim. do.

However, the technical problem to be achieved by the embodiments of the present application is not limited to the technical problems as described above, and other technical problems may exist.

As a technical means for achieving the above technical problem, the artificial intelligence-based claim claims and fraud pattern detection apparatus according to an embodiment of the present application, data pre-processing unit for deriving the characteristic variable by formulating the insurance claim details data, the feature An outlier detection unit that detects a normal pattern and an invalid pattern from the detected abnormal clusters based on the abnormal detection algorithm using a variable as an input, and performs an unsupervised learning using the normal pattern and the invalid pattern as inputs, thereby creating a new pattern of invalid claims. New pattern classification unit for detecting a new pattern detection unit, a new pattern detection unit for detecting a pattern of a new invalid claim by building a pattern discrimination model based on the supervised learning that inputs the new pattern of the invalid claim and the desired insurance claim pattern Supervised detection based on supervised learning based on variable and normal billing data The model may include an unjust claim determination unit that determines whether the new claim details are unfair.

According to an embodiment of the present application, the claim history data, and further comprising a database for recording at least one of the characteristic variable and the new pattern, the claim history data is billing data, contract data, payment data, insurance agent It may include at least one of data, customer data.

According to an embodiment of the present application, the outlier detection unit detects the normal value and the outlier value of each characteristic variable of the normal transaction and the invalid transaction of the insurance claim based on the abnormality detection algorithm using the characteristic variable as an input. The anomaly cluster can be detected based on the frequency of the corresponding feature variable.

According to the exemplary embodiment of the present application, the new pattern classification unit clusters the normal pattern and the invalid pattern based on a clustering algorithm, and detects the new pattern of the invalid claim based on the separation degree between the normal pattern and the invalid pattern. can do.

According to an embodiment of the present application, the new pattern detection unit may detect the pattern of the new invalid claim based on a classification algorithm.

Artificial intelligence-based claims fraud detection method according to an embodiment of the present invention, (a) formulating the claim details data to derive the characteristic variable, (b) based on the abnormality detection algorithm as the characteristic variable input Detecting a normal pattern and an invalid pattern from the detected abnormal cluster, (c) performing unsupervised learning with the normal pattern and an invalid pattern as input, and detecting a new pattern of an invalid claim, (d) the invalid pattern Constructing a pattern learning model based on supervised learning with input of new patterns of claims and pre-sorted insurance claim patterns to detect patterns of new fraudulent claims; and (e) Maps using the characteristic variables and normal claim data as inputs. It may include the step of determining whether the new claim details are unfair by building a learning-based fraud detection model.

According to one embodiment of the present application, the artificial intelligence-based claim fraud detection method further comprises the step of recording at least one of the claim history data, the characteristic variable and the new pattern, the claim history data is billing data , Contract data, payment data, insurance company data, and customer data.

According to an embodiment of the present application, the step (b) detects the normal value and the abnormal value of each characteristic variable of the normal transaction and the invalid transaction of the insurance claim based on the abnormality detection algorithm which takes the characteristic variable as an input, and The abnormal cluster can be detected based on the frequency of the feature variable corresponding to the outlier.

According to an embodiment of the present application, the step (c) clusters the normal pattern and the invalid pattern based on a clustering algorithm, and based on the degree of separation between the clusters of the normal pattern and the invalid pattern, a new pattern of the invalid claim is made. Can be detected.

According to an embodiment of the present application, step (d) may detect the pattern of the new invalid claim based on a classification algorithm.

The above-mentioned means for solving the problems are merely exemplary, and should not be construed as limiting the present application. In addition to the above-described exemplary embodiments, additional embodiments may exist in the drawings and detailed description of the invention.

According to the above-described problem solving means of the present application, it is possible to provide an artificial intelligence-based claims and claims pattern detection apparatus and method that provides a discrimination model that can determine a new pattern of claims unfair claims.

According to the above-described problem solving means of the present invention, AI-based claims fraud and fraud pattern detection device that can learn a new claim data to analyze a new claim pattern, and determine the type of claim pattern for a new claim and It may provide a method.

According to the above-described problem solving means of the present application, it is possible to provide an artificial intelligence-based claim unfair claim and unfair pattern detection apparatus and method for learning the claim claim-related data to determine unfairness for a new claim.

1 is a diagram illustrating a configuration of an artificial intelligence-based insurance claim unfair claim and unfair pattern detection device according to an embodiment of the present application.
FIG. 2 is a diagram illustrating an example of characteristic variables of an artificial intelligence-based insurance claim unfair claim and unfair pattern detection apparatus according to an embodiment of the present disclosure.
FIG. 3 is a diagram illustrating patterns classified by an abnormality detection algorithm of an artificial intelligence-based insurance claim unfair claim and unfair pattern detection device according to an embodiment of the present disclosure.
4 is a diagram illustrating a detection result of an anomaly detection algorithm of an artificial intelligence-based insurance claim unfair claim and an unfair pattern detection device according to an embodiment of the present application.
FIG. 5 is a diagram illustrating algorithm-specific accuracy of an invalid claim detection model of an artificial intelligence-based claim claim and an invalid pattern detection device according to an embodiment of the present disclosure.
6 is a flowchart illustrating a method of artificial intelligence-based insurance claim unfair claim and unfair pattern detection method according to an embodiment of the present application.

DETAILED DESCRIPTION Hereinafter, exemplary embodiments of the present disclosure will be described in detail with reference to the accompanying drawings so that those skilled in the art may easily implement the present disclosure. As those skilled in the art would realize, the described embodiments may be modified in various different ways, all without departing from the spirit or scope of the present invention. In the drawings, parts irrelevant to the description are omitted for simplicity of explanation, and like reference numerals designate like parts throughout the specification.

Throughout this specification, when a part is "connected" to another part, this includes not only "directly connected" but also "electrically connected" with another element in between. do.

Throughout this specification, when a member is said to be located on another member "on", "upper", "top", "bottom", "bottom", "bottom", this means that any member This includes not only the contact but also the case where another member exists between the two members.

Throughout this specification, when a part is said to "include" a certain component, it means that it can further include other components, without excluding the other components unless otherwise stated.

1 is a diagram illustrating a configuration of an artificial intelligence-based insurance claim unfair claim and unfair pattern detection device according to an embodiment of the present application.

Referring to FIG. 1, the AI-based insurance claim fraud and unfair pattern detection apparatus 100 includes a data preprocessor 110, an outlier detection unit 120, a new pattern classification unit 130, and a new pattern detection unit 140. And an unfair billing determination unit 150. The data preprocessor 110 may derive the characteristic variable by formulating the insurance claim history data.

FIG. 2 is a diagram illustrating an example of characteristic variables of an artificial intelligence-based insurance claim unfair claim and unfair pattern detection apparatus according to an embodiment of the present disclosure.

The description below will include not only claims unfair claims, but also claims, subsidies, security deposits, etc., but will be described based on claims unfair claims for convenience of explanation. For example, the insurance claim data may include at least one of claim data, insurance or subsidy-related contract data, insurance payment data, insurance company data, customer data, and insurance fraud detection result data. The data preprocessor 110 may derive the characteristic variable by formalizing the insurance claim data. For example, in the case of the customer data, the standardization of the insurance claim data means to quantify the income level of the customer, the number of hospitals visited by the customer, the number of disease reasons that the customer applied for, and the like. In addition, the characteristic variable derived by the data preprocessing unit 110 is a variable that may have a numerical value associated with the claim, for example, the customer ID number, whether the insurance fraud, the number of duplicate applications in the same bottle name, The number of months signed, the maximum number of contracts per day, the number of securities approved for payment, the number of securities applied for payment, the number of securities entered by the customer, the number of claims for insurance coverage, the change in credit rating, the type of contract signed, the total number of hospital visits, The number of reasons that the client applied for, the number of doctors the client met, the number of hospitals the client visited, the total number of days of effective admission / hospitalization, the number of treatment subjects, the level of the client's income, the number of changes in the Financial Planner, the number of losses, It may include at least one of the number of fraudulent FP contract.

FIG. 3 is a diagram illustrating patterns classified by an abnormality detection algorithm of an artificial intelligence-based insurance claim unfair claim and unfair pattern detection device according to an embodiment of the present disclosure.

The outlier detection unit 120 may detect a normal pattern and an invalid pattern from the detected abnormal cluster based on the abnormality detection algorithm using the characteristic variable as an input. The abnormality detection algorithm refers to an algorithm that detects a small number of abnormal data that does not fit the average characteristic of the entire data, that is, the insurance claim history data. Specifically, the outlier detection unit 120 detects the normal value and the outlier value of each characteristic variable of the normal transaction and the invalid transaction of the insurance claim based on the abnormality detection algorithm that takes the characteristic variable as an input, and the characteristic variable corresponding to the abnormal value. The abnormal clusters can be detected based on the frequency of. For example, the abnormality detection algorithm may include an isolation forest algorithm, and the isolation forest algorithm may detect an abnormal cluster by isolating abnormal data on a tree basis.

The outlier detection unit 120 may detect a small number of clusters detected by the abnormality detection algorithm as the abnormality cluster. In the case of insurance claim, since the ratio of unfair claims is smaller than that of normal claims, the above algorithm can isolate the abnormal data. That is, the outlier detection unit 120 may detect an abnormal cluster in an invalid pattern and detect a cluster that does not belong to the abnormal cluster in a normal pattern. Referring to FIG. 3, the outlier detection unit 120 may be classified into a normal pattern, a attention pattern, a suspicion pattern, and an abnormal pattern according to the isolation degree by the isolation forest algorithm. Attention and suspicion patterns may be fraudulent, but they may fall within the normal category with patterns not discriminated against.

4 is a diagram illustrating a detection result of an anomaly detection algorithm of an artificial intelligence-based insurance claim unfair claim and an unfair pattern detection device according to an embodiment of the present application.

Referring to FIG. 4, the outlier detection unit 120 may detect the normal value and the outlier value of the normal transaction and the unfair transaction of the claim through the abnormality detection algorithm. Unfair trades in claims can be detected as normal when there is an existing unfair trade pattern, while outliers are likely to be a new pattern of unfair trade. By way of example, the accuracy of normal and outlier detection using the isolation forest algorithm represents 87% performance.

The new pattern classification unit 130 may perform unsupervised learning using a normal pattern and an invalid pattern as an input to detect a new pattern of an invalid claim, that is, an invalid pattern. Unsupervised learning refers to an algorithm that learns by analyzing or clustering data itself, rather than constructing learning data. Since this is a known matter, a detailed description thereof will be omitted. The new pattern classification unit 130 may cluster the normal pattern and the invalid pattern based on a clustering algorithm, and detect the new pattern (unfair pattern) of the invalid claim based on the separation degree between the normal pattern and the invalid pattern. have.

For example, a logistic regression algorithm, a random forest algorithm, a support vector machine (SVM) algorithm, a decision algorithm, and a clustering algorithm may be used as the clustering algorithm for unsupervised learning. In addition to the above-described algorithm, the new pattern classification unit 130 may use clustering algorithms such as Extra Tree algorithm, XG Boost algorithm, Deep Learning algorithm, K-means clustering algorithm, Self-Organizing-Maps (SOM) algorithm, EM & Canopy algorithm. Can lead to unsupervised learning. The random forest algorithm is an algorithm that a lot of decision trees form a forest and averaging each prediction result into one result variable. The SVM algorithm determines the classification to which data belongs by dividing the largest boundary in the data distribution space. It is a non-probability algorithm. The Extra Tree algorithm is similar to the Random forest, but the algorithm is faster than the Random forest. The XGBoost algorithm is a boost algorithm that applies the result of the XGBoost tree to the next tree if the tree of the Random Forest is independent. Deep Learning algorithm is an algorithm that learns by adjusting the effect of the pattern of variable on the weight based on multi-layer Neural Network. In addition, K-means clustering algorithm is a traditional classification technique that repeats the target group into K clusters based on the mean value (similarity) of distance, and the SOM algorithm is based on the artificial neural network. It is a technique of learning and clustering. In addition, the EMI & Canopy algorithm refers to a technique for clustering by updating parameter values through the iterative process from the maximum probability to a given initial value.

The new pattern classification unit 130 may cluster the normal pattern and the invalid pattern into a plurality of pattern clusters through the clustering algorithm. That is, the pattern cluster may be a cluster of homogeneous feature variables. In addition, it is possible to detect the new pattern of the illegal claim based on the degree of separation between the groups of the pattern clusters, that is, the distance between the groups. The new pattern classification unit 130 may cluster similar normal patterns and invalid patterns, and classify dissimilar patterns into new patterns, and if it is determined that the pattern is a normal claim based on the characteristic variables of the new patterns, the normal patterns If it is determined that the new pattern is an invalid pattern, it may be detected as a new pattern of an invalid claim. Specifically, when the frequency of the feature variables belonging to the normal pattern or the wrong pattern is similar, the same to similar patterning (or identifier) may be given to similar feature variables. That is, when the patterning between the patterns is the same or similar, it can be said that the degree of separation between clusters is low. In addition, the claim pattern clustered at similar frequencies between the feature variables may be determined to be a clustered claim pattern, that is, a normal claim pattern rather than an unfair claim, based on the frequency of the feature variables resulting from the normal insurance claim.

On the other hand, in the case of a pattern grouped with feature variables having similar frequencies and feature variables having different frequencies, the labeling may be divided into different labeling. Such a pattern may be a new pattern of fraudulent claims, since the pattern is different from the frequency of the feature variables of the normal claim pattern (e.g., when the number of FP changes is relatively high compared to the feature variables of the normal claim pattern). It may be detected by the pattern classifier 130. Detecting a new pattern of fraudulent claims is an important factor for detecting a newly appearing fraudulent claim even in a limited human resource. By accumulating such a new pattern, the reliability of detection of an invalid claim can be improved, and a foundation for detecting unknown unknown claims that will appear in the future can be provided based on the accumulated pattern data.

The new pattern detection unit 140 may detect a pattern of the new invalid claim by constructing a pattern discrimination model based on supervised learning that takes a new pattern of an invalid claim and a pre-classified insurance claim pattern. Supervised learning means learning a model using pre-built training data. In addition, the new pattern detector 140 may detect the pattern of the new invalid claim based on a classification algorithm. The classification algorithm includes a decision algorithm and the decision algorithm may be, for example, a decision tree algorithm, but is not limited thereto. The new pattern detection unit 140 may learn a new pattern discrimination rule based on a decision algorithm that inputs a new pattern of unfair claims and a pre-classified insurance claim pattern. In addition, the new pattern detection unit 140 may build a pattern discrimination model of a new invalid claim including a new pattern discrimination rule. The above-described new pattern classification based on unsupervised learning can simply classify the new pattern based on the frequency, but it is not known by which variable. Therefore, new pattern discrimination rules can be learned based on supervised learning-based decision algorithms.

The new pattern discrimination rule means a rule capable of discriminating a claim pattern as a new pattern based on a characteristic variable of a threshold value or more. In addition, the new pattern discrimination rule may be integrated into a pattern discrimination model for new unfair claims. For example, referring to FIG. 4, even in the case of a normal transaction, an outlier may be detected, and it may not be said that there is no possibility that it is a new pattern of unfair claims. Therefore, the new pattern discrimination rule can be learned by inputting all the patterns including the new pattern of the unfair claim and the surviving claim pattern. In addition, the learning of the new pattern discrimination rule may be considered the importance of the feature variable. That is, among the various characteristic variables, the characteristic variables with high risk of fraudulent claims may have a relatively high importance, and the importance of these characteristic variables may be considered when learning new pattern discrimination rules. For example, the total number of significant hospital visits among the characteristic variables may be given a higher importance than the number of doctors the client met. That is, the new pattern detection unit 140 may construct a more accurate pattern discrimination model of new unfair claims by considering the importance of the feature variables.

FIG. 5 is a diagram illustrating algorithm-specific accuracy of an invalid claim detection model of an artificial intelligence-based claim claim and an invalid pattern detection device according to an embodiment of the present disclosure.

The invalid claim determination unit 150 may determine whether the new insurance claim is unfair by constructing a supervised detection model based on supervised learning using characteristic variables and normal billing data as inputs. The fraud detection model may output the fraud detection result by inputting the characteristic variable and the normal billing data. The invalid bill determination unit 150 may build the invalid bill detection model based on a classification / prediction algorithm based on supervised learning. The classification / prediction algorithm may be an algorithm described in the clustering algorithm described above, and thus redundant description thereof will be omitted.

The fraud detection model may output the fraud detection result for the new insurance details by inputting the characteristic variable and the normal billing data. As an example, the fraud detection model may output the fraudulent sedentary result based on the similarity between the feature variable and the normal billing data. If the similarity is less than a preset threshold, the insurance claim associated with the characteristic variable may be determined to be normal. If the similarity is more than the preset threshold, the insurance claim associated with the characteristic variable may be determined to be an invalid claim. The similarity may be calculated by at least one of K-means clustering algorithm, Self-Organizing-Maps (SOM) algorithm, EM & Canopy algorithm. The invalidity determination unit 150 may improve the accuracy of the invalidation detection model by repeatedly deriving the output by the input.

The invalid claim determination unit 150 may determine whether the new insurance claim is invalid based on the invalid claim detection model. When the new claim details are input to the constructed claim detection model, the wrong claim determination unit 150 may determine whether the new claim details are unfair based on the similarity as described above. At this time, the new claim details are also input to the pattern discrimination model constructed by the new pattern detection unit 140, so as to determine whether the new claim details are unjust claims and the pattern of the new claim, as well as the new claims. If the insurance claim is an invalid claim, it is possible to determine whether the claim is a new pattern.

According to an embodiment of the present application, the artificial intelligence-based claim fraud detection apparatus 100 may further include a database, and the database may record at least one of claims history data, the characteristic variable, and the new pattern. .

6 is a flowchart illustrating a method of artificial intelligence-based insurance claim unfair claim and unfair pattern detection method according to an embodiment of the present application.

The artificial intelligence-based insurance claim unfair claim and unfair pattern detection method illustrated in FIG. 6 may be performed by the artificial intelligence-based insurance claim unfair claim and unfair pattern detection apparatus described above with reference to FIGS. 1 to 5. Therefore, even if omitted below, the descriptions of the apparatus for analyzing new unfair claims through FIGS. 1 to 5 may be equally applicable to FIG. 6.

Referring to FIG. 6, in operation S610, the data preprocessor 110 may derive characteristic variables by shaping insurance claim details data. For example, in the case of the customer data, the standardization of the insurance claim data means to quantify the income level of the customer, the number of hospitals visited by the customer, the number of disease reasons that the customer applied for, and the like. In addition, the characteristic variable derived by the data preprocessing unit 110 is a variable that may have a numerical value associated with the claim, for example, the customer ID number, whether the insurance fraud, the number of duplicate applications in the same bottle name, The number of months signed, the maximum number of contracts per day, the number of securities approved for payment, the number of securities applied for payment, the number of securities entered by the customer, the number of claims for insurance coverage, the change in credit rating, the type of contract signed, the total number of hospital visits, The number of reasons that the client applied for, the number of doctors the client met, the number of hospitals the client visited, the total number of days of effective admission / hospitalization, the number of treatment subjects, the level of the client's income, the number of changes in the Financial Planner, the number of losses, It may include at least one of the number of fraudulent FP contract.

In operation S620, the outlier detection unit 120 may detect a normal pattern and an invalid pattern from the detected abnormal cluster based on the abnormality detection algorithm using the characteristic variable as an input. The abnormality detection algorithm refers to an algorithm that detects a small number of abnormal data that does not fit the average characteristic of the entire data, that is, the insurance claim history data. Specifically, the outlier detection unit 120 detects the normal value and the outlier value of each characteristic variable of the normal transaction and the invalid transaction of the insurance claim based on the abnormality detection algorithm that takes the characteristic variable as an input, and the characteristic variable corresponding to the abnormal value. The abnormal clusters can be detected based on the frequency of. For example, the abnormality detection algorithm may include an isolation forest algorithm, and the isolation forest algorithm may detect an abnormal cluster by isolating abnormal data on a tree basis. The outlier detection unit 120 may detect a small number of clusters detected by the abnormality detection algorithm as the abnormality cluster. In the case of insurance claim, since the ratio of unfair claims is smaller than that of normal claims, the above algorithm can isolate the abnormal data. That is, the outlier detection unit 120 may detect an abnormal cluster in an invalid pattern and detect a cluster that does not belong to the abnormal cluster in a normal pattern.

In operation S630, the new pattern classification unit 130 may perform unsupervised learning using the normal pattern and the invalid pattern as inputs to detect the new pattern of the invalid claim. Unsupervised learning refers to an algorithm that learns by analyzing or clustering data itself, rather than constructing learning data. Since this is a known matter, a detailed description thereof will be omitted. The new pattern classification unit 130 may cluster the normal pattern and the invalid pattern based on a clustering algorithm, and detect the new pattern of the invalid claim based on the separation degree between the normal pattern and the invalid pattern.

The new pattern classification unit 130 may cluster the normal pattern and the invalid pattern into a plurality of pattern clusters through the clustering algorithm. That is, the pattern cluster may be a cluster of homogeneous feature variables. In addition, it is possible to detect the new pattern of the illegal claim based on the degree of separation between the groups of the pattern clusters, that is, the distance between the groups. The new pattern classification unit 130 may cluster similar normal patterns and invalid patterns, and classify dissimilar patterns into new patterns, and if it is determined that the pattern is a normal claim based on the characteristic variables of the new patterns, the normal patterns If it is determined that the new pattern is an invalid pattern, it may be detected as a new pattern of an invalid claim. Specifically, when the frequency of the feature variables belonging to the normal pattern or the wrong pattern is similar, the same to similar patterning (or identifier) may be given to similar feature variables. That is, when the patterning between the patterns is the same or similar, it can be said that the degree of separation between clusters is low. In addition, the claim pattern clustered at similar frequencies between the feature variables may be determined to be a clustered claim pattern, that is, a normal claim pattern rather than an unfair claim, based on the frequency of the feature variables resulting from the normal insurance claim.

In operation S640, the new pattern detection unit 140 may detect a pattern of the new invalid claim by constructing a pattern discrimination model based on a supervised learning using the new pattern of the invalid claim and the pre-classified insurance claim pattern. Supervised learning means learning a model using pre-built training data. In addition, the new pattern detector 140 may detect the pattern of the new invalid claim based on a classification algorithm. The classification algorithm includes a decision algorithm and the decision algorithm may be, for example, a decision tree algorithm, but is not limited thereto. The new pattern detection unit 140 may learn a new pattern discrimination rule based on a decision algorithm that inputs a new pattern of unfair claims and a pre-classified insurance claim pattern. In addition, the new pattern detection unit 140 may build a pattern discrimination model of a new invalid claim including a new pattern discrimination rule. The above-described new pattern classification based on unsupervised learning can simply classify the new pattern based on the frequency, but it is not known by which variable. Therefore, new pattern discrimination rules can be learned based on supervised learning-based decision algorithms.

The new pattern discrimination rule means a rule capable of discriminating a claim pattern as a new pattern based on a characteristic variable of a threshold value or more. In addition, the new pattern discrimination rule may be integrated into a pattern discrimination model for new unfair claims. For example, an outlier may be detected even in a normal transaction, and it may not be said that it is not likely to be a new pattern of fraudulent claims. Therefore, the new pattern discrimination rule can be learned by inputting all the patterns including the new pattern of the unfair claim and the surviving claim pattern. In addition, the learning of the new pattern discrimination rule may be considered the importance of the feature variable. That is, among the various characteristic variables, the characteristic variables with high risk of fraudulent claims may have a relatively high importance, and the importance of these characteristic variables may be considered when learning new pattern discrimination rules. For example, the total number of significant hospital visits among the characteristic variables may be given a higher importance than the number of doctors the client met. That is, the new pattern detector 140 may construct a more accurate pattern discrimination model of new unfair claims by considering the importance of the feature variables.

In operation S650, the invalid claim determination unit 150 may determine whether the new insurance claim is invalid by constructing a supervised detection model based on supervised learning that uses characteristic variables and normal billing data as inputs. The fraud detection model may output the fraud detection result by inputting the characteristic variable and the normal billing data. The invalid bill determination unit 150 may build the invalid bill detection model based on a classification / prediction algorithm based on supervised learning. The classification / prediction algorithm may be an algorithm described in the clustering algorithm described above, and thus redundant description thereof will be omitted.

The fraud detection model may output the fraud detection result for the new insurance details by inputting the characteristic variable and the normal billing data. As an example, the fraud detection model may output the fraudulent sedentary result based on the similarity between the feature variable and the normal billing data. If the similarity is less than a preset threshold, the insurance claim associated with the characteristic variable may be determined to be normal. If the similarity is more than the preset threshold, the insurance claim associated with the characteristic variable may be determined to be an invalid claim.

The invalid claim determination unit 150 may determine whether the new insurance claim is invalid based on the invalid claim detection model. When the new claim details are input to the constructed claim detection model, the wrong claim determination unit 150 may determine whether the new claim details are unfair based on the similarity as described above. At this time, the new claim details are also input to the pattern discrimination model constructed by the new pattern detection unit 140, so as to determine whether the new claim details are unjust claims and the pattern of the new claim, as well as the new claims. If the insurance claim is an invalid claim, it is possible to determine whether the claim is a new pattern.

According to an embodiment of the present disclosure, the AI-based claim fraud detection method may further include recording at least one of claims detail data, the characteristic variable, and the new pattern.

According to an embodiment of the present disclosure, an artificial intelligence-based claim fraud detection method may be implemented in a program instruction form that may be executed by various computer means and recorded in a computer readable medium. The computer readable medium may include program instructions, data files, data structures, and the like, alone or in combination. The program instructions recorded on the media may be those specially designed and constructed for the present invention, or they may be of the kind well-known and available to those having skill in the computer software arts. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tape, optical media such as CD-ROMs, DVDs, and magnetic disks, such as floppy disks. Magneto-optical media, and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, flash memory, and the like. Examples of program instructions include not only machine code generated by a compiler, but also high-level language code that can be executed by a computer using an interpreter or the like. The hardware device described above may be configured to operate as one or more software modules to perform the operations of the present invention, and vice versa.

The above description of the present application is intended for illustration, and it will be understood by those skilled in the art that the present invention may be easily modified in other specific forms without changing the technical spirit or essential features of the present application. Therefore, it should be understood that the embodiments described above are exemplary in all respects and not restrictive. For example, each component described as a single type may be implemented in a distributed manner, and similarly, components described as distributed may be implemented in a combined form.

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 equivalents should be construed as being included in the scope of the present application.

100: AI-based claims fraud detection device
110: data preprocessor
120: outlier detection unit
130: new pattern classification unit
140: new pattern detection unit
150: unfair claim judgment

Claims (11)

In the artificial intelligence-based claim unfair claim and unfair pattern detection device,
A data preprocessor for deriving characteristic variables by formulating insurance claim data;
An outlier detection unit that detects a normal pattern and an invalid pattern from the detected abnormal cluster based on the abnormality detection algorithm using the characteristic variable as an input;
A new pattern classification unit for performing unsupervised learning using the normal pattern and the unfair pattern as an input to detect a new pattern of unfair claims;
A new pattern detector configured to detect a pattern of a new invalid claim by constructing a pattern discrimination model based on supervised learning that takes the new pattern of the invalid claim and a pre-classified insurance claim pattern; And
Unfair claim determination unit to build a supervised detection model based on supervised learning based on the characteristic variable and the normal claim data to determine whether the new claim details unfair,
Artificial intelligence-based claims fraud detection device comprising a. The method of claim 1,
And a database for recording at least one of the claim history data, the feature variable, and the new pattern.
The claim details data includes at least one of claim data, contract data, payment data, insurance company data, customer data, artificial intelligence-based claims fraud detection device. The method of claim 1,
The outlier detection unit,
Based on the abnormality detection algorithm using the characteristic variable as an input, the normal value and the abnormal value of each characteristic variable of the normal transaction and the invalid transaction of the insurance claim are detected, and the abnormality cluster is based on the frequency of the characteristic variable corresponding to the abnormal value. The AI-based claim fraud detection device that is to detect. The method of claim 1,
The new pattern classification unit,
And claiming the new pattern of the illegal claim based on a clustering algorithm based on a clustering algorithm and detecting the new pattern of the illegal pattern based on the degree of separation between the normal pattern and the invalid pattern. The method of claim 1,
The new pattern detector,
And claim new pattern of claim fraudulent detection based on a classification algorithm. In the artificial intelligence-based insurance claim detection method,
(a) formulating claim details data to derive feature variables;
(b) detecting a normal pattern and an invalid pattern from the detected abnormal clusters based on the abnormal detection algorithm using the characteristic variables as inputs;
(c) performing unsupervised learning using the normal pattern and the invalid pattern as an input to detect a new pattern of the invalid claim;
(d) detecting a pattern of a new fraudulent claim by constructing a pattern learning model based on supervised learning using the new pattern of the fraudulent claim and a pre-sorted insurance claim pattern; And
(e) constructing an unsupervised detection model based on supervised learning using the characteristic variables and normal billing data as inputs, and determining whether the new insurance claims are unfair;
Artificial intelligence-based insurance fraud detection method comprising a. The method of claim 6,
The AI-based claim fraud detection method may further include recording at least one of the claim details data, the characteristic variable, and the new pattern.
The claim details data includes at least one of billing data, contract data, payment data, insurance company data, customer data, artificial intelligence-based claims fraud detection method. The method of claim 6,
In step (b),
Based on the abnormality detection algorithm using the characteristic variable as an input, the normal value and the abnormal value of each characteristic variable of the normal transaction and the invalid transaction of the insurance claim are detected, and the abnormality cluster is based on the frequency of the characteristic variable corresponding to the abnormal value. AI-based claims fraud detection method that is to detect. The method of claim 6,
Step (c) is,
And collecting the normal pattern and the invalid pattern based on a clustering algorithm, and detecting the new pattern of the invalid claim based on the degree of separation between the normal pattern and the invalid pattern. The method of claim 6,
In step (d),
And detecting a pattern of the new invalid claim based on a classification algorithm. A computer-readable recording medium having recorded thereon a program for executing the method of any one of claims 6 to 10.

Images