JP7009658B2 - Malicious delinquency detection device and method based on artificial intelligence - Google Patents

Description

The present application relates to a malicious delinquency detection device and method based on artificial intelligence.

The existing insurance fraud prevention system is a method of deriving rules based on the experience and knowledge of the examiner for insurance cases claimed as the basis of business rules, and is divided into survey target rules and non-survey target rules. It is calculated. However, as insurance fraud becomes more intelligent and sophisticated, new fraudulent claims patterns or fraud patterns require continuous updates, and there have been developments of fraudulent claims and fraudulent pattern detection systems.

In addition, various fraud problems or malicious customer problems are occurring not only in insurance companies but also in various industries, and each company should leave the cost leakage without taking any measures against such problems. Nor can we spend an infinite amount of research costs for the purpose of completely revealing the problem. Therefore, each company needs to determine the level of research efforts to an appropriate level so as not to spend excessively large research costs while reducing unnecessary expenses from the above-mentioned problems that may occur in the field.

The technique behind the present application is disclosed as Patent Document 1.

Korean Registered Patent Gazette No. 10-0862181

The present application is for solving the above-mentioned problems of the prior art, and is an artificial intelligence capable of learning malicious delinquency data, analyzing a new malicious delinquency pattern, and determining the type of malicious delinquency pattern for a new case. The purpose is to provide an intelligence-based malicious delinquency detection device and method.

However, the technical problem to be achieved by the embodiment of the present application is not limited to the above-mentioned technical problem, and other technical problems may exist.

As a technical means for achieving the above technical problems, the malicious delinquency detection device based on the artificial intelligence according to the embodiment of the present application has a data preprocessing unit that calculates characteristic variables from rental fee payment breakdown data and the characteristics. A pattern detector that detects normal patterns and delinquency patterns based on anomaly detection algorithms that input variables, and a new pattern for rental fee payment based on a non-teaching learning platform crowd algorithm that inputs the normal patterns and delinquency patterns. Includes a new pattern classification unit that classifies new patterns, and a new payment pattern detection unit that constructs a pattern discrimination model that inputs the normal pattern, the delinquency pattern, and the new pattern, and detects the malicious pattern of the new malicious delinquent. be able to.

According to one embodiment of the present application, the rental fee payment breakdown data further includes a database for recording at least one of the rental fee payment breakdown data, the characteristic variable, and the new pattern, and the rental fee payment breakdown data includes a business partner code and whether or not there is a delinquency. , Corporate classification, bond transfer status, business closure classification, years of operation, total number of contracts, number of past due history, maximum number of months overdue, etc. can be included at least one of them.

According to one embodiment of the present application, the pattern detection unit detects normal values and abnormal values from characteristic variables corresponding to normal payment and delinquency of rental fee payment based on the abnormality detection algorithm, and responds to the abnormal values. Abnormal communities can be detected based on the frequency of characteristic variables.

According to one embodiment of the present application, the pattern detection unit can detect the normal pattern and the delinquency pattern based on the abnormal community.

According to one embodiment of the present application, the new pattern classification unit crowds the normal pattern and the delinquent pattern based on the crowd algorithm, and pays the rental fee based on the degree of separation between the normal pattern and the delinquent pattern. New patterns can be categorized.

According to one embodiment of the present application, the new payment pattern detection unit constructs a pattern discrimination model based on a classification algorithm of a teaching learning platform that inputs the normal pattern, the delinquency pattern, and the new pattern, and the maliciousness. The pattern can be detected.

According to one embodiment of the present application, a malicious delinquent detection model of a teaching and learning platform that inputs the characteristic variable and the rental fee payment data is constructed to predict whether or not the customer is a malicious delinquent. It can further include a delinquent predictor.

According to one embodiment of the present application, the malicious delinquent predictor can calculate the detection result of whether or not the person is a malicious delinquent based on the degree of similarity between the characteristic variable and the normally paid data. ..

The malicious delinquency detection method of the artificial intelligence platform according to the embodiment of the present application is normal based on (a) a step of calculating a characteristic variable from the rental fee payment breakdown data and (b) an abnormality detection algorithm using the characteristic variable as an input. The stage of detecting the pattern and the delinquency pattern, (c) the stage of classifying the new pattern of rental fee payment based on the crowd algorithm of the non-teaching learning platform that inputs the normal pattern and the delinquency pattern, and (d) the stage of the normality. It is possible to include a step of constructing a pattern discrimination model in which a pattern, the delinquency pattern and the new pattern are input, and detecting a malicious pattern of a new malicious delinquent person.

According to one embodiment of the present application, the malicious delinquency detection method of the artificial intelligence platform further includes the step of recording at least one of the rental fee payment breakdown data, the characteristic variable, and the new pattern, and the rental fee payment. Breakdown data may include at least one of data such as customer code, delinquency status, corporate classification, bond transfer status, business closure category, years of operation, total number of contracts, delinquency history, and maximum delinquency months. can.

According to one embodiment of the present application, in the step (b), the normal value and the abnormal value are detected from the characteristic variables corresponding to the normal payment and the delinquency of the rental fee payment based on the abnormality detection algorithm, and the abnormal value is set to the abnormal value. Abnormal communities can be detected based on the frequency of the corresponding characteristic variables.

According to one embodiment of the present application, step (b) can detect the normal pattern and the delinquency pattern based on the abnormal community.

According to one embodiment of the present application, the step (c) crowds the normal pattern and the delinquent pattern based on the crowd algorithm, and pays the rental fee based on the degree of separation between the normal pattern and the delinquent pattern. New patterns can be categorized.

According to one embodiment of the present application, in the step (d), a pattern discrimination model is constructed based on a classification algorithm of a teaching learning platform that inputs the normal pattern, the delinquency pattern, and the new pattern, and the malicious pattern is described. Can be detected.

According to one embodiment of the present application, the malicious delinquency detection method of the artificial intelligence platform is as follows: (e) A customer can construct a malicious delinquency detection model of the instruction learning platform by inputting the characteristic variable and the rental fee payment data. It can further include a step of predicting whether or not it is a malicious delinquent.

According to one embodiment of the present application, the step (e) can calculate the detection result of whether or not the person is a malicious delinquent based on the similarity between the characteristic variable and the normally paid data.

The problem-solving means described above are merely exemplary and should not be construed with the intent of limiting the present application. In addition to the exemplary examples described above, there may be additional examples in the drawings and detailed description of the invention.

According to the above-mentioned problem-solving means of the present application, an artificial intelligence-based malicious delinquency detection device and method capable of learning malicious delinquency data, analyzing a new malicious delinquency pattern, and discriminating a malicious delinquency pattern type for a new case. Can be provided.

It is a drawing which illustrated the structure of the malicious delinquency detection device of the artificial function base by one Embodiment of this application.

It is a figure which illustrated the example of the correlation between the characteristic variables of the malicious delinquency detection device of the artificial intelligence base by one Example of this application.

It is a figure which illustrated the detection result of the abnormality detection algorithm of the malicious delinquency detection device of the artificial intelligence base by one Embodiment of this application.

It is a drawing which illustrated the pattern classified by the crowd algorithm of the malicious delinquency detection device of the artificial intelligence base by one Embodiment of this application.

It is a figure which illustrated the performance by algorithm of the malicious delinquent person detection model of the malicious delinquency detection device of the artificial intelligence base by one Embodiment of this application.

It is a drawing which illustrated the pattern detection rule example of the malicious delinquency detection device of the artificial intelligence base by one Embodiment of this application.

It is a drawing which illustrated the pattern detection rule example of the malicious delinquency detection device of the artificial intelligence base by one Embodiment of this application.

It is a drawing which illustrated the pattern detection rule example of the malicious delinquency detection device of the artificial intelligence base by one Embodiment of this application.

It is a drawing which illustrated the flow of the malicious delinquency detection method of the artificial intelligence base by one Embodiment of this application.

Hereinafter, examples of the present application will be described in detail so that a person having ordinary knowledge in the technical field to which the present application belongs can easily carry out the procedure with reference to the attached drawings. However, the present application can be embodied in a variety of different forms and is not limited to the embodiments described herein. Then, in order to clearly explain the present application in the drawings, parts unrelated to the description are omitted, and similar parts are designated by similar drawing reference numerals throughout the specification.

In the entire specification of the present application, if one part is "connected" to another part, this is not only when it is "directly connected", but also when another element is sandwiched between them and "electrically". Including the case of being "concatenated to".

Assuming that, in the entire specification of the present application, one member is located "above", "at the top", "at the top", "below", "at the bottom", "at the bottom", this is This includes not only the case where one member is in contact with another member but also the case where another member is present between the two members.

In the whole specification of the present application, when a certain component is "included", this may include other components, rather than excluding other components, unless otherwise specified. It means that you can do it.

FIG. 1 is a drawing illustrating the configuration of a malicious delinquency detection device based on an artificial function according to an embodiment of the present application.

Referring to FIG. 1, the artificial intelligence-based malicious delinquency detection device 100 includes a data preprocessing unit 110, a pattern detection unit 120, a new pattern classification unit 130, a new payment pattern detection unit 140, a malicious delinquency person prediction unit 150, and a database. 160 can be included.

It is obvious that the explanations described below include not only malicious delinquency of rental costs but also malicious delinquency of leasing costs, loans, redemptions, etc., but for convenience of explanation, focusing on malicious delinquency of rental costs. explain. The rental fee payment breakdown data is, for example, at least one of data such as customer code, delinquency status, corporate classification, bond transfer status, closed business classification, operating years, total number of contracts, delinquency history, and maximum delinquency months. Can include one or one.

FIG. 2 is a drawing illustrating an example of the correlation between the characteristic variables of the malicious delinquency detection device of the artificial intelligence base according to the embodiment of the present application.

The data pre-processing unit 110 can calculate characteristic variables from the rental fee payment breakdown data. Specifically, the data pre-processing unit 110 can formulate the rental fee payment breakdown data and derive characteristic variables. The standardization of rental fee payment breakdown data means, for example, quantifying and categorizing the closed business classification, operating years, total number of contracts, normal contract history, number of progressed contracts, and the like. Further, the characteristic variable derived by the data preprocessing unit 110 is a variable that can have a quantified value in connection with the payment of rental expenses, for example, whether or not there is delinquency, whether or not there is a corporate classification, whether or not there is a bond transfer, and whether or not the business is closed. Classification, number of years of operation, total number of contracts, number of normal contracts, number of progressed contracts, maximum number of months overdue, maximum number of extensions, total billing amount, total storage amount, total delinquency amount, presence / absence of negative response of general management, general management It can include at least one of the number of call histories and the corresponding month data. Referring to FIG. 2, it can be said that the higher the correlation between the characteristic variables, the higher the probability that the pattern detection unit 120 will detect it as a delinquency pattern. The database 160 can record at least one of the rental fee payment breakdown data, the characteristic variable, and the new pattern.

FIG. 3 is a drawing illustrating the detection result of the abnormality detection algorithm of the malicious delinquency detection device of the artificial intelligence base according to the embodiment of the present application.

The pattern detection unit 120 can detect a normal pattern and a delinquency pattern based on an anomaly detection algorithm that inputs a characteristic variable. The anomaly detection algorithm means an algorithm for detecting a small number of anomalous data that do not match the average characteristics of the entire data, that is, the rental fee payment breakdown data.

The pattern detection unit 120 can detect normal values and abnormal values from characteristic variables corresponding to normal payment and malicious delinquency of rental fee payment patterns based on an abnormality detection algorithm. In addition, the pattern detection unit 120 can detect an abnormal community based on the frequency of characteristic variables corresponding to abnormal values. By way of example, the anomaly detection algorithm can include an Isolation Forest algorithm, which can detect anomalous communities by isolating anomalous data on a tree basis. By way of example, the accuracy of normal and outlier detection using the isolation forest algorithm represents 89% performance.

Specifically, the pattern detection unit 120 can detect a small number of communities detected by the anomaly detection algorithm as anomalous communities. In addition, the pattern detection unit 120 can detect a normal pattern and a delinquent pattern based on the abnormal community. In the case of rental fee payment, the rate of delinquency is smaller than the rate of normal payment, so abnormal data can be isolated through the abnormality detection algorithm. That is, the pattern detection unit 120 can detect an abnormal community as a delinquent pattern and a community that does not belong to the abnormal community as a normal pattern. Referring to FIG. 3, the pattern detection unit 120 can detect the normal value and the abnormal value of the rental fee payment pattern for normal payment and malicious delinquency, respectively, through the abnormality detection algorithm. Specifically, the malicious delinquency of rental fee payment can be detected as a normal value in the case of an existing delinquency pattern, while the abnormal value of malicious delinquency is likely to be a new pattern of malicious delinquency.

FIG. 4 is a drawing illustrating a pattern classified by a crowd algorithm of a malicious delinquency detection device based on an artificial intelligence according to an embodiment of the present application.

FIG. 4 (a) illustrates a community of normal patterns, and FIG. 4 (b) illustrates a crowd of abnormal patterns. Referring to FIG. 4, the new pattern classification unit 130 can classify a new pattern of rental fee payment based on a crowd algorithm of a non-teaching learning platform that inputs the normal pattern and the delinquency pattern. Non-teaching learning does not construct learning data, but means an algorithm for learning while analyzing or crowding the data itself. Since this is a publicly known matter, a specific description thereof will be omitted. The new pattern classification unit 130 can crowd the normal pattern and the delinquency pattern based on the crowd algorithm. Further, the new pattern classification unit 130 can classify the new pattern for payment of the rental fee based on the degree of separation between the communities of the normal pattern and the delinquent pattern.

As an example, as the crowd algorithm for non-teaching learning, a K-Neightbors algorithm, a random forest algorithm, a RadientBosting algorithm, an SVM (Support Vector Machine) algorithm, a decision-making algorithm, and a crowd algorithm can be used. In addition to the above-mentioned algorithms, the new pattern classification unit 130 includes an Extra Tree algorithm, an XG Boost algorithm, a Deep Learning algorithm, a K-means clustering algorithm, an SOM (Self-Organizing-Maps) algorithm, and an EM & Canon algorithm. Non-teaching learning can be performed through various crowd algorithms. The Random Forest algorithm is an algorithm in which a large number of Decision Trees compose a Forest and averages each prediction result with one result variable, and the SVM algorithm divides the boundary of the widest width in the data distribution space and divides the data. It is a non-probabilistic algorithm that determines the classification to which it belongs. The Extra Tree algorithm is similar to the Random forest, but faster than the Random forest, and the XGBoost algorithm applies the results of the Random Forest Tree to the next tree, independent of the Random Forest Tree. It is a boss method algorithm. The Deep Learning algorithm is an algorithm that learns while adjusting the influence of the pattern of variables on the result with a weighted value based on the multi-layered Neural Network. The K-means clustering algorithm is a technique of repeatedly subdividing a target group into K communities based on the mean value (similarity) of distances by a traditional classification technique, and the SOM algorithm is a technique of repeatedly subdividing an artificial neural network. It is a technique to learn the input pattern of the training set by weighted value and cluster it as a basis. The EM & Canopy algorithm also means a technique for updating and crowding parameter values through an iterative process from the given initial value with the greatest potential.

The new pattern classification unit 130 can subdivide the normal pattern and the delinquent pattern into a plurality of pattern communities through the crowd algorithm. A pattern community is a community of homogeneous characteristic variables that separate based on the degree of separation between communities, i.e., the distance between communities. Through this, the normal pattern, the delinquency pattern, and the new pattern can be detected. Specifically, when the frequencies of the characteristic variables belonging to the normal pattern or the delinquency pattern are similar, the same or similar patterning (or identification character) can be given to the similar characteristic variables. In this way, when the patterning between patterns is the same or similar, it can be said that the degree of separation between the communities is low. The new pattern classification unit 130 can crowd similar normal patterns and delinquent patterns, and classifies dissimilar patterns as new patterns. If it is determined that the pattern is similar to normal payment based on the characteristic variable, it can be detected as a normal pattern, and if it is determined that the new pattern is a similar delinquency pattern, it can be detected as a malicious delinquency pattern.

On the other hand, in the case of a pattern crowded with characteristic variables having dissimilar frequencies, it can be classified by labeling different from the above labeling. Since such a pattern is different from the frequency of characteristic variables of a normal similar payment pattern, it can be a new pattern of rental fee payment and can be detected by the new pattern classification unit 130.

As an example, referring to (a) of FIG. 4 above, an abnormal value can be detected even in the case of normal payment. This means that it may be classified as an overdue pattern as a new pattern of rental fee payment. Detecting new patterns of rental payments is an important factor in detecting newly emerging malicious delinquents, even in situations where human resources are limited. By accumulating such new patterns, the reliability of malicious delinquent detection can be improved, and new unknown malicious delinquent cases that will appear in the future will be detected based on the accumulated pattern data. A possible base can be provided.

The new payment pattern detection unit 140 can construct a pattern discrimination model that inputs the normal pattern, the delinquency pattern, and the new pattern, and can detect the malicious pattern of the new malicious delinquent. Specifically, the new payment pattern detection unit 140 constructs a pattern discrimination model based on the classification algorithm of the instruction learning platform that inputs the normal pattern, the delinquency pattern, and the new pattern, and detects the malicious pattern. be able to. Teaching learning means learning a model by utilizing pre-constructed training data (training data). That is, the new payment pattern detection unit 140 can detect the pattern of the malicious delinquent, that is, the malicious pattern based on the classification algorithm. The classification algorithm includes, but is not limited to, a decision-making algorithm, and the decision-making algorithm can be, for example, a Decision Tree algorithm.

Further, the new payment pattern detection unit 140 can learn the pattern detection rule based on the decision-making algorithm that inputs the new pattern of rental fee payment and the classified rental fee payment pattern. In addition, the new payment pattern detection unit 140 can build a pattern discrimination model for malicious delinquents including pattern detection rules. In the above-mentioned new pattern classification of the non-teaching learning platform, the new pattern can be classified simply based on the frequency, but it is unknown by what variable. However, the decision-making algorithms of the teaching and learning infrastructure can confirm this through pattern detection rules.

The pattern detection rule means a rule that can determine a rental fee payment pattern based on a characteristic variable equal to or higher than a threshold value. Such a pattern detection rule can be learned by inputting all patterns including a new pattern of rental fee payment and a classified rental fee payment pattern. Furthermore, the pattern detection rules can be integrated and constructed as a pattern discrimination model for malicious delinquents. On the other hand, in the learning of pattern detection rules, the importance of characteristic variables can be taken into consideration. That is, among various characteristic variables, characteristic variables with a high risk of malicious delinquency can have a relatively high importance, and the importance of such characteristic variables is taken into consideration when learning the pattern detection rule. be able to. That is, the new payment pattern detection unit 140 can construct a more accurate pattern discrimination model for malicious delinquents by considering the importance of the characteristic variable.

FIG. 5 is a drawing illustrating the performance of the malicious delinquent detection device of the malicious delinquency detection device based on the artificial intelligence base by algorithm according to the embodiment of the present application, and FIGS. 6a to 6c show the artificial intelligence according to the embodiment of the present application. It is a drawing which illustrated the example of the pattern detection rule of the malicious delinquency detection device of the base.

The malicious delinquent person prediction unit 150 constructs a malicious delinquent person detection model of a teaching and learning platform that inputs the characteristic variable and the rental fee payment data, and predicts whether or not the customer is a malicious delinquent person. Can be done. The malicious delinquent detection model can output the malicious delinquent detection result by inputting the characteristic variable and rental fee payment data. The malicious delinquent prediction unit 150 can construct the malicious delinquent detection model based on the classification / prediction algorithm of the teaching learning platform. Since the classification / prediction algorithm can be the algorithm described in the crowd algorithm described above, duplicate description will be omitted.

Further, the malicious delinquent person prediction unit 150 can calculate the detection result of whether or not the person is a malicious delinquent person based on the degree of similarity between the characteristic variable and the normal payment data. The malicious delinquent detection model can output the detection result of whether or not the customer corresponds to the malicious delinquent by inputting the characteristic variable and the rental fee payment data. As an example, the malicious delinquent detection model can output the detection result of whether or not it corresponds to the malicious delinquent based on the similarity between the characteristic variable and the normally paid data. If the similarity is less than the preset threshold value, the rental fee payment pattern linked with the characteristic variable is judged to be normal, and if the similarity is equal to or higher than the preset threshold value, the rental fee payment pattern is linked with the characteristic variable. The rental fee payment pattern can be determined as a malicious delinquent. The similarity can be calculated by at least one of a K-means clustering algorithm, a SOM (Self-Organizing-Maps) algorithm, and an EM & Canony algorithm. The malicious delinquent person prediction unit 150 can improve the accuracy of the malicious delinquent person detection model by repeatedly deriving the output by the input.

The malicious delinquent person prediction unit 150 can predict whether or not the customer is a malicious delinquent person based on the malicious delinquent person detection model. If a new rental fee payment breakdown is input to the constructed malicious delinquent detection model, the malicious delinquent prediction unit 150 determines whether or not the customer is a malicious delinquent based on the similarity as described above. Can be predicted. At this time, by inputting the new rental cost payment breakdown into the pattern discrimination model constructed by the new payment pattern detection unit 140, whether or not the new rental cost payment breakdown is malicious overdue and the corresponding new rental cost payment breakdown. In addition to grasping the pattern of, if the breakdown of new rental fee payment is malicious delinquency, it is possible to determine whether or not there is a new pattern of malicious delinquency.

FIG. 7 is a drawing illustrating the flow of the malicious delinquency detection method of the artificial intelligence base according to the embodiment of the present application.

The malicious delinquency detection method of the artificial intelligence base shown in FIG. 7 can be performed by the malicious delinquency detection device 100 of the artificial intelligence base described through FIGS. 1 to 6c described above. Therefore, even if the content is omitted below, the content described for the malicious delinquency detection device 100 based on the artificial intelligence through FIGS. 1 to 6c can be similarly applied to FIG. 7.

Referring to FIG. 7, in step S710, the data preprocessing unit 110 can calculate the characteristic variable from the rental fee payment breakdown data. Specifically, the data pre-processing unit 110 can formulate the rental fee payment breakdown data and derive characteristic variables. The standardization of rental fee payment breakdown data means, for example, quantifying and categorizing the closed business classification, operating years, total number of contracts, normal contract history, number of progressed contracts, and the like. Further, the characteristic variable derived by the data preprocessing unit 110 is a variable that can have a quantified value in connection with the payment of rental expenses, for example, whether or not there is delinquency, whether or not there is a corporate classification, whether or not there is a bond transfer, and whether or not the business is closed. Classification, number of years of operation, total number of contracts, number of normal contracts, number of progressed contracts, maximum number of months overdue, maximum number of extensions, total billing amount, total storage amount, total delinquency amount, presence / absence of negative response of general management, general management It can include at least one of the number of call histories and the corresponding month data. It can be said that the higher the correlation between the characteristic variables, the higher the probability that the pattern detection unit 120 will detect it as a delinquent pattern. The database 160 can record at least one of the rental fee payment breakdown data, the characteristic variable, and the new pattern.

In step S720, the pattern detection unit 120 can detect a normal pattern and a delinquency pattern based on an anomaly detection algorithm that inputs a characteristic variable. The anomaly detection algorithm means an algorithm for detecting a small number of anomalous data that do not conform to the average characteristics of the entire data, that is, the rental fee payment breakdown data.

The pattern detection unit 120 can detect normal values and abnormal values from characteristic variables corresponding to normal payment and malicious delinquency of rental fee payment patterns based on an abnormality detection algorithm. In addition, the pattern detection unit 120 can detect an abnormal community based on the frequency of characteristic variables corresponding to abnormal values. By way of example, the anomaly detection algorithm can include an Isolation Forest algorithm, which can detect anomalous communities by isolating anomalous data on a tree basis. As an example, the accuracy of normal and outlier detection using the isolation forest algorithm represents 89% performance.

Specifically, the pattern detection unit 120 can detect a small number of communities detected by the anomaly detection algorithm as anomalous communities. In addition, the pattern detection unit 120 can detect a normal pattern and a delinquent pattern based on the abnormal community. In the case of rental fee payment, the rate of delinquency is smaller than the rate of normal payment, so abnormal data can be isolated through the abnormality detection algorithm. That is, the pattern detection unit 120 can detect an abnormal community as a delinquent pattern and a community that does not belong to the abnormal community as a normal pattern. The pattern detection unit 120 can detect the normal value and the abnormal value of the rental fee payment pattern for normal payment and malicious delinquency through the abnormality detection algorithm. Specifically, the malicious delinquency of rental fee payment can be detected as a normal value in the case of an existing delinquency pattern, while the abnormal value of malicious delinquency is likely to be a new pattern of malicious delinquency.

In step S730, the new pattern classification unit 130 can classify the new pattern of rental fee payment based on the crowd algorithm of the non-teaching learning platform that inputs the normal pattern and the delinquency pattern. Non-teaching learning does not construct learning data, but means an algorithm for learning while analyzing or crowding the data itself. Since this is a publicly known matter, a specific description thereof will be omitted. The new pattern classification unit 130 can crowd the normal pattern and the delinquency pattern based on the crowd algorithm. Further, the new pattern classification unit 130 can classify the new pattern for payment of the rental fee based on the degree of separation between the communities of the normal pattern and the delinquent pattern.

The new pattern classification unit 130 can subdivide the normal pattern and the delinquent pattern into a plurality of pattern communities through the crowd algorithm. A pattern community is a community of homogeneous characteristic variables that separate based on the degree of separation between communities, i.e., the distance between communities. Through this, the normal pattern, the delinquency pattern, and the new pattern can be detected. Specifically, when the frequencies of the characteristic variables belonging to the normal pattern or the delinquency pattern are similar, the same or similar patterning (or identification character) can be given to the similar characteristic variables. In this way, when the patterning between patterns is the same or similar, it can be said that the degree of separation between communities is low. The new pattern classification unit 130 can crowd similar normal patterns and delinquent patterns, and classifies dissimilar patterns as new patterns. If it is determined that the pattern is similar to normal payment based on the characteristic variable, it can be detected as a normal pattern, and if it is determined that the new pattern is a similar delinquency pattern, it can be detected as a malicious delinquency pattern.

On the other hand, in the case of a pattern crowded with characteristic variables having dissimilar frequencies, it can be classified into labeling different from the above labeling. Since such a pattern is different from the frequency of characteristic variables of a normal similar payment pattern, it can be a new pattern of rental fee payment and can be detected by the new pattern classification unit 130. As an example, an abnormal value can be detected even in the case of normal payment. This means that it may be classified as an overdue pattern as a new pattern of rental fee payment. Detecting new patterns of rental payments is an important factor in detecting newly emerging malicious delinquents, even in situations where human resources are limited. By accumulating such new patterns, the reliability of malicious delinquent detection can be improved, and even unknown new malicious delinquent cases that will appear in the future can be detected based on the accumulated pattern data. A foundation that can be provided can be provided.

The new payment pattern detection unit 140 can construct a pattern discrimination model that inputs the normal pattern, the delinquency pattern, and the new pattern, and can detect the malicious pattern of the new malicious delinquent. Specifically, the new payment pattern detection unit 140 constructs a pattern discrimination model based on the classification algorithm of the instruction learning platform that inputs the normal pattern, the delinquency pattern, and the new pattern, and detects the malicious pattern. be able to. Teaching learning means learning a model by utilizing pre-constructed training data (training data). That is, the new payment pattern detection unit 140 can detect the pattern of the malicious delinquent, that is, the malicious pattern based on the classification algorithm. The classification algorithm includes, but is not limited to, a decision-making algorithm, and the decision-making algorithm can be, for example, a Decision Tree algorithm.

Further, the new payment pattern detection unit 140 can learn the pattern detection rule based on the decision-making algorithm that inputs the new pattern of rental fee payment and the classified rental fee payment pattern. In addition, the new payment pattern detection unit 140 can build a pattern discrimination model for malicious delinquents including pattern detection rules. In the above-mentioned new pattern classification of the non-teaching learning platform, the new pattern can be classified simply based on the frequency, but it is unknown by what variable. However, the decision-making algorithms of the teaching and learning infrastructure can confirm this through pattern detection rules.

The pattern detection rule means a rule that can determine a rental fee payment pattern based on a characteristic variable equal to or higher than a threshold value. Such a pattern detection rule can be learned by inputting all patterns including a new pattern of rental fee payment and a classified rental fee payment pattern. Furthermore, the pattern detection rules can be integrated and constructed as a pattern discrimination model for malicious delinquents. On the other hand, in the learning of pattern detection rules, the importance of characteristic variables can be taken into consideration. That is, among various characteristic variables, characteristic variables with a high risk of malicious delinquency can have a relatively high importance, and the importance of such characteristic variables is taken into consideration when learning the pattern detection rule. be able to. That is, the new payment pattern detection unit 140 can construct a more accurate pattern discrimination model for malicious delinquents by considering the importance of the characteristic variable.

In step S740, the malicious delinquent predictor 150 constructs a malicious delinquent detection model of a teaching and learning platform using the characteristic variable and the rental fee payment data as input, and determines whether or not the customer is a malicious delinquent. Can be predicted. The malicious delinquent detection model can output the malicious delinquent detection result by inputting the characteristic variable and rental fee payment data. The malicious delinquent prediction unit 150 can construct the malicious delinquent detection model based on the classification / prediction algorithm of the teaching learning platform. Since the classification / prediction algorithm can be the algorithm described in the crowd algorithm described above, duplicate description will be omitted.

Further, the malicious delinquent person prediction unit 150 can calculate the detection result of whether or not the person is a malicious delinquent person based on the degree of similarity between the characteristic variable and the normal payment data. The malicious delinquent detection model can output the detection result of whether or not the customer corresponds to the malicious delinquent by inputting the characteristic variable and the rental fee payment data. As an example, the malicious delinquent detection model can output the detection result of whether or not it corresponds to the malicious delinquent based on the similarity between the characteristic variable and the normally paid data. If the similarity is less than the preset threshold value, the rental fee payment pattern linked with the characteristic variable is judged to be normal, and if the similarity is equal to or higher than the preset threshold value, the rental fee payment pattern is linked with the characteristic variable. The rental fee payment pattern that has been made can be determined as a malicious delinquent. The similarity can be calculated by at least one of a K-means clustering algorithm, a SOM (Self-Organizing-Maps) algorithm, and an EM & Canony algorithm. The malicious delinquent prediction unit 150 can improve the accuracy of the malicious delinquent detection model by repeatedly deriving the output by the input.

The malicious delinquent person prediction unit 150 can predict whether or not the customer is a malicious delinquent person based on the malicious delinquent person detection model. If a new rental fee payment breakdown is input to the constructed malicious delinquent detection model, the malicious delinquent prediction unit 150 determines whether or not the customer is a malicious delinquent based on the similarity as described above. Can be predicted. At this time, by inputting the new rental cost payment breakdown into the pattern discrimination model constructed by the new payment pattern detection unit 140, whether or not the new rental cost payment breakdown is malicious overdue and the corresponding new rental cost payment breakdown. In addition to grasping the pattern of, if the breakdown of new rental fee payment is malicious delinquency, it is possible to determine whether or not it is a new pattern of malicious delinquency.

The malicious delinquency detection method of the artificial intelligence platform according to the embodiment of the present application is embodied in a program instruction form that can be performed through various computer means, and can be recorded on a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, etc. alone or in combination. The program instructions recorded on the medium may be those specially designed and configured for the present invention, or those known and usable by those skilled in the art of computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks and magnetic tapes, optical media such as CD-ROMs and DVDs, and floptic discs. ), And hardware devices specially configured to store and perform program instructions such as ROM, RAM, flash memory, etc. included. Examples of program instructions include not only machine language code as created 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 can be configured to operate as one or more software modules to perform the operation of the present invention, and vice versa.

The above description of the present application is for illustration purposes only, and a person having ordinary knowledge of the technical field to which the present application belongs may change to another concrete form without changing the technical idea or essential features of the present application. You can see that it is easily deformable. Therefore, it should be understood that the examples described above are exemplary in all respects and are not limiting. For example, each component described in the single type may be carried out in a distributed manner, and the components described in the same distributed form may also be carried out in a combined form.

The scope of the present application is represented by the scope of claims described later rather than the above detailed description, and the meaning and scope of the claims and all modified or modified forms derived from the concept of equality thereof are the scope of the present application. Should be interpreted as being included in.

Claims (9)

In a malicious delinquency detector based on artificial intelligence,
A data pre-processing unit that calculates characteristic variables from rental fee payment breakdown data,
A pattern detection unit that detects a normal pattern and a delinquency pattern based on an anomaly detection algorithm that inputs the characteristic variable, and
A new pattern classification unit that classifies new patterns for rental fee payment based on the crowd algorithm of the non-teaching learning platform that inputs the normal pattern and the delinquency pattern, and
A new payment pattern detector that detects a malicious pattern of a new malicious delinquent by constructing a pattern discrimination model that inputs the normal pattern, the delinquency pattern, and the new pattern.
Including
The pattern detection unit is
Based on the anomaly detection algorithm, normal and abnormal values of normal payment and malignant delinquency are detected from the characteristic variables corresponding to normal payment and delinquency of rental fee payment.
The abnormal community is detected based on the frequency of the characteristic variable corresponding to the abnormal value, and the normal pattern and the delinquency pattern are detected based on the abnormal community.
The new pattern classification unit is
The normal pattern and the delinquent pattern are crowded based on the crowd algorithm, and the new pattern of the rental fee payment is classified based on the degree of separation between the normal pattern and the delinquent pattern.
A pattern discrimination model is constructed based on the classification algorithm of the teaching and learning platform that inputs the normal pattern, the delinquency pattern, and the new pattern, and the malicious pattern is detected.
The pattern discrimination model includes a pattern detection rule learned based on a decision-making alcoholism that inputs a new pattern of rental fee payment and a classified rental fee payment pattern, and is an artificial intelligence-based malicious delinquency detection device. .. Further including a database recording at least one of the rental fee payment breakdown data, the characteristic variable and the new pattern.
The rental fee payment breakdown data is at least one of data such as customer code, delinquency status, corporate classification, bond transfer status, suspension of business classification, number of years of operation, total number of contracts, number of delinquency history, and maximum number of delinquent months. The malicious delinquency detection device of the artificial intelligence base according to claim 1, which comprises. It further includes a malicious delinquent predictor that predicts whether or not a customer is a malicious delinquent by constructing a malicious delinquent detection model of a teaching and learning platform that inputs the characteristic variable and rental fee payment breakdown data. The malicious delinquency detection device based on the artificial intelligence according to claim 1 or 2 , which is an object. The malicious delinquent prediction unit
The artificial intelligence-based malicious delinquency detection device according to claim 3 , which calculates the detection result of whether or not the person is a malicious delinquent based on the similarity between the characteristic variable and the normal payment data. In the malicious delinquency detection method of artificial intelligence base,
(A) At the stage of calculating characteristic variables from rental fee payment breakdown data,
(B) A step of detecting a normal pattern and a delinquency pattern based on an anomaly detection algorithm using the characteristic variable as an input, and
(C) The stage of classifying a new pattern of rental fee payment based on the crowd algorithm of the non-teaching learning platform that inputs the normal pattern and the delinquency pattern, and
(D) A step of constructing a pattern discrimination model using the normal pattern, the delinquency pattern, and the new pattern as inputs to detect a malicious pattern of a new malicious delinquent.
Including
The step (b) is
Based on the anomaly detection algorithm, normal and abnormal values of normal payment and malignant delinquency are detected from the characteristic variables corresponding to normal payment and delinquency of rental fee payment.
The abnormal community is detected based on the frequency of the characteristic variable corresponding to the abnormal value, and the normal pattern and the delinquency pattern are detected based on the abnormal community.
The step (c) is
The normal pattern and the delinquent pattern are crowded based on the crowd algorithm, and the new pattern of the rental fee payment is classified based on the degree of separation between the normal pattern and the delinquent pattern.
A pattern discrimination model is constructed based on the classification algorithm of the teaching and learning platform that inputs the normal pattern, the delinquency pattern, and the new pattern, and the malicious pattern is detected.
The pattern discrimination model includes a pattern detection rule learned based on a decision-making alcoholism that inputs a new pattern of rental fee payment and a classified rental fee payment pattern, and is a malicious delinquency detection method based on artificial intelligence. .. Further including the step of recording at least one of the rental fee payment breakdown data, the characteristic variable and the new pattern.
The rental fee payment breakdown data is at least one of data such as customer code, delinquency status, corporate classification, bond transfer status, suspension of business classification, number of years of operation, total number of contracts, number of delinquency history, and maximum number of delinquent months. The malicious delinquency detection method of the artificial intelligence base according to claim 5 , which comprises. (E) It further includes a step of constructing a malicious delinquent detection model of a teaching and learning platform using the characteristic variable and the rental fee payment breakdown data as input, and predicting whether or not the customer is a malicious delinquent. , The malicious delinquency detection method of the artificial intelligence base according to claim 5 or 6 . The step (e) is
The method for detecting malicious delinquency based on artificial intelligence according to claim 7 , wherein the detection result of whether or not the person is a malicious delinquent is calculated based on the degree of similarity between the characteristic variable and the normal payment data. A program for executing the method according to any one of claims 5 to 8 on a computer.

Images