JP4755911B2 - Scoring system and scoring method for calculating fraud score using credit card - Google Patents

Description

  The present invention relates to a scoring system and a scoring method for calculating a score that reflects the probability that a credit card is illegally used.

  When using a credit card, check the credit balance from the store that accepted the use of the card to the card company, for example, in order to prevent others who have picked up the card from impersonating themselves In addition, inquiries about the possibility of unauthorized use are being made. In an unauthorized use inquiry, a scoring model that calculates the possibility of unauthorized use as a score is generally used (see, for example, Patent Documents 1 to 4).

  As a method of scoring model, item values corresponding to various items such as time, amount, store, etc. are extracted from authorization data received when using a credit card, and in the range that approximates each item value in past authorization data A method of calculating an unauthorized use score by calculating the probability of occurrence of fraud is used. In such a scoring model, corresponding to the item value of each item, data on the number of authorization data in a predetermined range and the number of cases that correspond to unauthorized use are accumulated, and the appearance of fraud from these data Calculate the probability and score the possibility of fraud.

  A problem in the above scoring system is weighting for each item in calculating the score. In order to appropriately reflect the influence of the fraud appearance probability in each item in the score indicating the possibility of fraud, it is possible to apply logistic regression analysis.

  Another possibility is that if the item value corresponding to each item is associated with a sample of past authorization data, it is necessary to set a certain range for the item value and collect a sample to obtain a sufficient sample parameter. Since it is not possible to secure the score, the reliability of the score is lowered. Therefore, the item values for each item are categorized and samples are collected, and the probability of fraudulent appearance in each category is calculated. However, since the category itself is not a real number, score directly. Cannot be used to calculate As a technology that can cope with such issues, by converting item values to logit values in credit risk judgment, samples can be distributed in a linear model, the categories can be made real, and a highly accurate model can be generated. An invention relating to a possible credit risk model determination device is disclosed (see Patent Document 5).

JP 2004-334526 A JP 2004-334527 A JP 2004-334644 A JP 2004-348536 A JP 2004-334737 A

  When the invention described in Patent Document 5 is used, it is possible to calculate a score using a scoring model in which samples are appropriately allocated and the categories are realized. Therefore, the default probability is calculated using only the data received for the loan applicant as the object of the determination. On the other hand, in the field of the credit card targeted by the present invention, since the user continuously uses the credit card, not only the authorization data regarding each use but also the last use situation is taken into consideration. As a result, it is possible to make a more accurate determination as in an example in which an unusual appearance probability is high because an unusual usage pattern continues.

  The present invention has been made to cope with such a problem, and is a scoring system and a scoring method for calculating a score reflecting the probability that the use of a credit card is illegal. It is an object of the present invention to provide a scoring system and a scoring method adapted to the nature of a credit card that a category is converted into a real number and appropriately reflected in a score and is continuously used by a user.

The present invention that solves such a problem is a scoring system for calculating a score that reflects the probability that credit card usage is fraudulent, and that stores authorization data received every time a credit card is used. Data storage means, history data storage means for storing, for each user, history data relating to past credit card usage history of a credit card user, and items included in the authorization data stored in the authorization data storage means For the predetermined item included in the category of the item value and the authorization data stored in the authorization data storage means, the item value of the item included in the authorization data and the history data stored in the history data storage means category field values of the new item indicating a change in the item value of the item contained A logit value conversion table for storing logit values calculated from fraudulent appearance ratios in which fraud has occurred by category in each item in association with each category for each item, and an authorization stored in the authorization data storage means items included in the data, and per predetermined item included in the authorization data stored in the authorization data storage means, the item value of the item included in the authorization data, the history data stored in the history data storage means for a new item indicating a change in the item value of the item included, the contribution to the score of each item were calculated by logistic regression analysis, a regression coefficient table that associates and stores the respective items, the use of credit card Authorization data that accepts authorization data for the request And with means, an item value specifying means for specifying the item value of the predetermined item which the authorization data receiving unit is included in the authorization data received, the identifying the category corresponding to the item value the item value specifying means is specified 1 category specifying means, history data acquiring means for reading out and acquiring history data corresponding to the user specified from the authorization data received by the authorization data receiving means from the history data storage means, and the authorization data receiving for a given item that means is included in the authorization data received, a new indicating the item value of the item included in the authorization data, the change of the item value of the item included in the history data which the history data acquisition unit acquires Item value calculating means for calculating the item value of the item, and corresponding to the item value calculated by the item value calculating means Obtaining a second category specifying means for specifying that category, with reference to the logit value conversion table, the logit value first category identifying means corresponding to the identified category, is read out from the logit value conversion table The first logit value acquisition means that performs the first reference to the regression coefficient table, and the regression coefficient corresponding to the item for which the first category specification means has specified the category is read out from the regression coefficient table and acquired . The logarithmic coefficient acquisition means and the logit value conversion table, the logit value corresponding to the category specified by the second category specifying means is read out from the logit value conversion table and acquired. Referring to the acquisition means and the regression coefficient table, the second category specifying means The regression coefficients corresponding to the item to identify over, a second regression coefficient obtaining means for obtaining by reading from the regression coefficient table were obtained in the first logit value acquisition means and the second logit value acquiring means Regression acquired by the first regression coefficient acquisition unit and the second regression coefficient acquisition unit for each item corresponding to the logit value in the logit value read and acquired from the logit value conversion table including the logit value A score calculation means for calculating a score that reflects the probability that the credit card use request is fraudulent from the sum of the values obtained by multiplying the regression coefficients obtained by reading from the regression coefficient table including the coefficients. It is a scoring system characterized by comprising.

In the present invention, in scoring for determining unauthorized use of a credit card, not only items included in the received authorization data but also items included in the authorization data and history data are subject to score accumulation. By adding an item generated from the items included in, a scoring system corresponding to the risk inherent in the credit card can be constructed. The items generated from the items included in the authorization data to be accumulated and the items included in the history data include, for example, the ratio between the current usage amount and the previous usage amount, the current product code and the previous product code match or mismatch, etc. Can be used.

Further, according to the present invention, with respect to the authorization data received by the authorization data receiving means , items of predetermined items included in the history data stored in the history data storage means corresponding to a credit card user relating to the authorization data. History data updating means for updating a value to an item value of the item included in the authorization data may be provided.

  With this configuration, it is possible to generate history data necessary for the next scoring while scoring the target authorization data.

Further, the present invention provides the logit value conversion table for each item value category of a new item calculated from item values of two or more items included in the authorization data stored in the authorization data storage means. The logit value calculated from the fraud appearance ratio in which fraud has appeared for each item value category in the item is stored in association with each category for each item, and the regression coefficient table is stored in the authorization data storage means For new items that calculate item values from the item values of two or more items included in the authorization data, the degree of contribution of each item to the score calculated by logistic regression analysis is stored in association with each item. Te, for the given item of which authorization data receiving unit is included in the authorization data received , Corresponding to the item value calculated by the second item value calculating means and the second item value calculating means for calculating the item value of the new item from the item values of two or more items included in the authorization data Referring to the third category specifying means for specifying the category and the logit value conversion table, the logit value corresponding to the category specified by the third category specifying means is read from the logit value conversion table and acquired. A third logit value acquisition unit and a regression coefficient table are referred to, and a regression coefficient corresponding to the item whose category is specified by the third category specifying unit is read from the regression coefficient table and acquired. Regression coefficient acquisition means, wherein the score calculation means adds the logistic value to the logit value acquired by the third logit value acquisition means. For items corresponding to the preparative value it may also be characterized by calculating the score using the value obtained by multiplying the obtained regression coefficient of the third regression coefficient acquisition unit.

With this configuration, the items to be scored are not only the items generated from the items included in the received authorization data, the items included in the authorization data, and the items included in the history data , but also the authorization data. By adding an item generated from the included items, it is possible to further increase the accuracy of scoring. For the item generated from the items included in the authorization data to be integrated, for example, the ratio of the current usage amount to the usage limit amount can be used.

In the logit value conversion table , each of the item value categories of new items calculated from item values of two or more items included in the history data stored in the history data storage means The logit value calculated from the fraud appearance ratio in which fraud has appeared for each item value category in the item is stored in association with each category for each item, and the regression coefficient table is stored in the history data storage means As for new items for calculating item values from item values of two or more items included in the history data, the degree of contribution to the score of each item calculated by logistic regression analysis is stored in association with each item. Te, from the item value of 2 or more items included in the history data to which the history data acquisition unit has acquired, the new items Refer to the third item value calculating means for calculating the eye value, the fourth category specifying means for specifying the category corresponding to the item value calculated by the third item value calculating means, and the logit value conversion table. The logit value corresponding to each category specified by the fourth category specifying means is read from the logit value conversion table and acquired, and referring to the regression coefficient table, A fourth regression coefficient acquisition unit that reads out and acquires a regression coefficient corresponding to each item for which the fourth category identification unit has identified a category from the regression coefficient table, and the score calculation unit includes: The fourth regression coefficient is acquired for each item corresponding to the logit value in the logit value acquired by the fourth logit value acquisition means. It may be characterized in that for calculating the score using the value obtained by multiplying the obtained regression coefficient stages.

According to this structure, the items to be accumulated score, not only the items generated from items included in the items and history data included in the items and authorization data included in the authorization data received, each user By adding an item included in the history data, scoring accuracy can be further improved. As items included in the history data to be accumulated, for example, data on the latest usage frequency and usage amount, and data on past usage trends such as day of the week and time of use can be used.

  Furthermore, this invention can also be specified as a scoring method using the scoring system concerning this invention.

That is, the scoring method according to the present invention is a scoring method for calculating a score reflecting the probability that the use of a credit card is illegal, and the computer system for calculating the score includes a credit card. An authorization data storage unit for storing authorization data received every time the card is used, and a history data storage unit for storing history data related to credit card users' past credit card usage history for each user. have been, the computer system, per predetermined item included the authorization item value categories included in the authorization data stored in the data storage unit, and the authorization data stored in the authorization data storage unit, authorization and item values of the items included in the data, the historical data rating The category of item value of the new item indicating a change in the item value of the item included in the stored historical data part, the logit value fraud by category is calculated from the occurrence of fraud ratio that appeared in each of the items, Storing in the logit value conversion table in association with each category for each item, and the computer system stored in the authorization data storage unit and the items included in the authorization data stored in the authorization data storage unit per predetermined item included in the authorization data, the item value of the item included in the authorization data, for a new item indicating a change in the item value of the item included in the history data stored in the history data storage unit, The contribution of each item to the score calculated by logistic regression analysis And storing in the regression coefficient table in association with each item, the logit value conversion table and the regression coefficient table data are updated, and a score reflecting the probability that the credit card is illegally used is calculated. In this case, the computer system receives an authorization data reception step for receiving authorization data relating to a credit card usage request, and the computer system receives the authorization data received in the authorization data reception step. and item value specifying step of specifying, said computer system, the first category specifying step of specifying a category corresponding to the item value specified in the Attribute value specifying step, the computer system, in the authorization data receiving step A history data acquisition step for reading out and acquiring history data corresponding to a user specified from the received authorization data from the history data storage unit, and the computer system includes the authorization data received in the authorization data receiving step. for a given items, items for calculating the item value of the item included in the authorization data, the item value of the new item indicating a change in the item value of the item included in the history data acquired by the history data acquiring step A value calculating step, a second category specifying step for specifying a category corresponding to the item value calculated in the item value calculating step by the computer system, and the computer system referring to the logit value conversion table, mosquitoes identified in the first category specifying step A first logit value obtaining step of obtaining a logit values read from the logit value conversion table corresponding to Gori, the computer system, by referring to the regression coefficient table, a category in the first category specifying step A first regression coefficient acquisition step of reading out and acquiring a regression coefficient corresponding to the identified item from the regression coefficient table, and the computer system refers to the logit value conversion table, and in the second category identification step A second logit value obtaining step of reading out and obtaining a logit value corresponding to the identified category from the logit value conversion table; and the computer system refers to the regression coefficient table and performs the second category specifying step. Corresponds to the item specified by That the regression coefficient, and the second regression coefficient acquisition step of acquiring read from the regression coefficient table, the computer system, the first logit value obtaining step and the second logit values logit value acquired at the acquisition step the logit value acquired by reading from the logit value conversion table containing, for each of the items corresponding to the logit value, the obtained regression coefficient in the first regression coefficient acquisition step and the second regression coefficient acquisition step A score calculation step of calculating a score that reflects the probability that the credit card use request is fraudulent from the sum of the values multiplied by the regression coefficient obtained by reading from the regression coefficient table including
It is a scoring method characterized by having.

Also, the scoring method according to the present invention, the computer system, the authorization data received by the authorization data receiving step, stored in the history data storage unit corresponding to the user's credit card according to the authorization data A history data update step of updating an item value of a predetermined item included in the history data to an item value of the item included in the authorization data may be provided.

Furthermore, in the scoring method according to the present invention, the logit value conversion table includes item values of new items calculated from item values of two or more items included in the authorization data stored in the authorization data storage unit. For each category, logit values calculated from fraud occurrence ratios where fraud appeared for each item value category in each item are stored in association with each category for each item, and the regression coefficient table stores the authorization data. Regarding new items for calculating item values from item values of two or more items included in the authorization data stored in the storage unit, the contribution of each item to the score calculated by logistic regression analysis is as follows. association with be stored, said computer system, said authorization data receiving step A second item value calculating step of calculating an item value of the new item from item values of two or more items included in the authorization data for a predetermined item included in the received authorization data; A third category specifying step for specifying a category corresponding to the item value calculated in the second item value calculating step; and the computer system refers to the logit value conversion table to determine the third category specifying step. A third logit value acquisition step of reading out and acquiring a logit value corresponding to the category specified in step 1 from the logit value conversion table; and the computer system refers to the regression coefficient table to specify the third category The regression coefficient corresponding to the item whose category was identified in the step is A third regression coefficient acquisition step that reads and acquires from a coefficient table, and in the score calculation step, the computer system adds the logit value acquired in the third logit value acquisition step to the logit value acquired in the third logit value acquisition step. For the item corresponding to the logit value, the score may be calculated using a value obtained by multiplying the regression coefficient acquired in the third regression coefficient acquisition step .

Furthermore, in the scoring method according to the present invention, the logit value conversion table includes item values of new items calculated from item values of two or more items included in the history data stored in the history data storage unit. For each category, logit values calculated from fraud occurrence ratios where fraud appeared for each item value category in each item are stored in association with each category for each item, and the regression coefficient table stores the history data. Regarding new items for calculating item values from item values of two or more items included in the history data stored in the storage unit, the contribution of each item to the score calculated by logistic regression analysis is as follows. association with be stored, historical data which the computer system was acquired by the history data acquiring step A third item value calculation step for calculating the item value of the new item from item values of two or more items included, and the item value calculated by the computer system in the third item value calculation step A fourth category specifying step for specifying a category to be performed, and the logit value conversion corresponding to the category specified in the fourth category specifying step by the computer system referring to the logit value conversion table. A fourth logit value acquisition step of reading out and acquiring from the table, and the computer system refers to the regression coefficient table, and sets the regression coefficient corresponding to the item whose category is specified in the fourth category specifying step, A fourth regression coefficient acquisition step of reading out and acquiring from the regression coefficient table; In the score calculation step, the computer system acquires the logit value acquired in the fourth logit value acquisition step and the item corresponding to the logit value in the fourth regression coefficient acquisition step. The score may be calculated using a value multiplied by a regression coefficient .

  According to the present invention, when calculating a score reflecting the probability that the credit card usage is illegal, the sample categories are converted into real numbers and appropriately reflected in the score, and the comparison with the usage tendency for each user is also reflected. By calculating the score, more accurate scoring is possible.

  The best mode for carrying out the present invention will be described below in detail with reference to the drawings and numerical examples. The following description shows an example of an embodiment of the scoring system according to the present invention, and the present invention is an example of items included in authorization data and history data used in the following description. It is not limited.

  FIG. 1 is a diagram showing an outline of a scoring system according to the present invention. FIG. 2 is a block diagram showing the configuration of the scoring system according to the present invention. FIG. 3 is a diagram illustrating an example of an illegal appearance probability for each usage time zone. FIG. 4 is a diagram illustrating a first example in which the illegal appearance probabilities for each usage time period are converted to logit values and rearranged. FIG. 5 is a diagram illustrating a second example in which the illegal appearance probabilities for each usage time period are converted to logit values and rearranged. FIG. 6 is a diagram showing an example of a logit value conversion table in the scoring system according to the present invention. FIG. 7 is a diagram showing an example of a regression coefficient table in the scoring system according to the present invention. FIG. 8 is a diagram showing an example of authorization data used in the scoring system according to the present invention. FIG. 9 is a diagram showing an example of history data used in the scoring system according to the present invention. FIG. 10 is a diagram showing an example of data in which item values for score calculation used in the scoring system according to the present invention are specified. FIG. 11 is a diagram showing an example of data in which item values for score calculation used in the scoring system according to the present invention are converted into logit values. FIG. 12 is a flowchart showing the flow of the scoring method according to the present invention.

  The scoring system according to the present invention is used as shown in FIG. A host computer 20 of a card company that issues a credit card is connected to a CAT terminal 30 or the like installed in a store of a member store through a settlement network. When a credit card member tries to pay with a credit card at a member store, the CAT terminal reads the credit card number, the product code of the product to be purchased, the usage amount, etc., and asks for the sales approval necessary to determine whether payment is possible. The authorization data is transmitted to the card company host computer 20.

  Upon receiving the authorization data, the card company host computer 20 confirms whether or not the usage amount included in the authorization data is within the range of the usage amount set for the credit card in order to determine whether or not the payment by the credit card is possible. . Furthermore, in order to prevent the credit card from being used by a malicious finder or the like, the scoring system 10 calculates a score indicating the possibility that the credit card is illegally used. Note that the scoring system 10 for calculating the score may be provided inside the card company, or may be provided outside a specialist or the like and connected via a network.

  When the scoring system 10 calculates the score, the card company host computer 20 that has received the score determines that there is a possibility of unauthorized use when the score exceeds a predetermined threshold, and the CAT terminal A message or the like that prompts the user to confirm the identity of the user is transmitted. On the side of the CAT terminal 30 that has received the request, the clerk at the member store asks the user for identity verification, thereby preventing unauthorized use by a malicious acquirer.

  The configuration of the scoring system according to the present invention will be described with reference to FIG. The scoring system 10 according to the present invention is connected to the card company host computer 20 so as to accept the authorization data to be scoring and respond with the calculated score. The configuration of the scoring system 10 is not particularly limited, and may be configured by either one computer or a plurality of computers, and constitutes a part of a computer having other functions. It may be a thing.

  The scoring system 10 includes a model analysis unit 13 that creates a model used for scoring, and a scoring unit 14 that calculates a score for each authorization data. The model analysis unit 13 and the scoring unit 14 include Both are specified as functions, and the hardware configuration for executing the necessary arithmetic processing is not particularly limited. For the logit value conversion analysis unit 131, the logistic regression analysis unit 132, the authorization data reception unit 141, the behavior data acquisition unit 142, the score calculation unit 143, and the behavior update unit 144, corresponding programs are read into the memory area of the computer. Each function is realized by executing predetermined arithmetic processing. In the logit value conversion table 133 and the regression coefficient table 134, data necessary for arithmetic processing is stored in a predetermined storage area allocated in the computer.

  The authorization DB 12 stores authorization data in which the authorization data itself previously received from the card company host computer 20 via the data transmission / reception unit 11 is added with information indicating unauthorized use and data such as gender. ing. The method for accumulating authorization data in the authorization DB 12 is not particularly limited, and the authorization data accepted for scoring may be stored for each acceptance, and once a day separately from scoring. For a period of time such as once a week, a certain period may be stored in a batch by batch processing.

  Note that, in the authorization data stored in the authorization DB 12, information indicating that if unauthorized use is recorded. Therefore, a predetermined range of item values of an item included in the authorization data is categorized, for example, “usage amount 5,000 yen to 10,000 yen”, “usage time 9 am to 10 am”, etc. When determined, it is possible to obtain fraud occurrence probabilities reflecting the results of samples corresponding to the category.

  The data stored in the authorization DB 12 is typical data in which predetermined information is added to the data received for each use, but the behavior item creation unit 151 uses the data (users) ) History data relating to each usage history is created and stored in a table or record provided for each user in the behavior DB 15. The accumulation method and update method of history data in the behavior DB 15 are not particularly limited, but initial data accumulation is performed by the operation of the behavior item creation unit 151, and the subsequent update is the operation of the behavior update unit 144 described later. It is possible to do by.

  The model analysis unit 13 has a function of analyzing a model necessary for scoring and creating a table for use in scoring. The logit value conversion analysis unit 131 categorizes the item values for the items included in the authorization data and the behavior items created from the past authorizations stored in the authorization DB 12, and calculates the incorrect appearance probability for each category. In order to be able to use the category as a real number, processing is performed to replace the item value with the logit value. Although this process will be described in detail later, the logit value corresponding to each item value is stored in association with the logit value conversion table 133 for the calculated result.

  Here, the item value replaced with the logit value is not limited to the value corresponding to the item included in the authorization data. Items generated from the items included in the authorization data (for example, items such as the ratio of the current usage amount to the usage limit), items included in the history data stored in the behavior DB 15, items included in the authorization data and history If item values corresponding to items generated from the items included in the data (for example, the ratio between the current usage amount and the previous usage amount, the current product code and the previous product code match or mismatch, etc.) are used, more accurate High scoring is possible.

  The logistic regression analysis unit 132 uses a logistic regression analysis to calculate a coefficient to be multiplied by the score calculated from the item value of each item in order to adjust the contribution to the score of each item for the items included in the authorization data. Calculate. As for the calculated result, the regression coefficient corresponding to each item is stored in association with the regression coefficient table 134.

  As for the items subject to logistic regression analysis, the items generated from the items included in the authorization data, the items included in the history data, the items included in the authorization data, and the items included in the history data as described above. If the generated items are to be subject to scoring, these items must also be subject to logistic regression analysis.

  The scoring unit 14 has a function of calculating a score by performing a predetermined calculation process on the authorization data received for scoring. The authorization data accepting unit 141 accepts authorization data to be subjected to scoring for sales approval from the card company host computer 20 via the data transmitting / receiving unit 11. The behavior data acquisition unit 142 acquires history data corresponding to the card user included in the received authorization data from the behavior DB 15.

  The score calculation unit 143 specifies item values for items used for score calculation included in the authorization data and items used for score calculation included in the history data. Here, it is also possible to calculate item values for items generated from items included in the authorization data, items generated from the items included in the authorization data, and items included in the history data. In this way, each logit value corresponding to the category to which the specified or calculated item value belongs is acquired with reference to the logit value conversion table 133. In addition, for the items for which the item values for score calculation are specified or calculated, the regression coefficient corresponding to each item is specified with reference to the regression coefficient table 134.

  The logit value for each item specified in this way is weighted with a regression coefficient, and the score is calculated based on the total value. The calculated score is returned to the card company host computer 20 via the data transmission / reception unit 11.

  In addition, the behavior update unit 144 reflects the usage corresponding to the authorization data received this time in the history data of the user whose score is to be calculated (for example, the data on the previous usage amount is updated to the current time). To update the history data. The updated history data is stored as the corresponding user history data in the behavior DB 15.

  The basic concept of score calculation in the scoring system according to the present invention will be described in detail below with reference to FIGS. The authorization data for which the score is calculated includes, for example, an item value indicating the usage time for the item of the usage time of the card that has received the authorization data. Using this item value as a reference, the degree of unauthorized appearance probability is referred to from past authorization data samples and reflected in the score.

  FIG. 3 shows the distribution with the usage time on the horizontal axis and the illegal appearance probability for each usage time on the vertical axis. In order to calculate a highly accurate score, it is preferable to increase the number of samples as much as possible so that the fraud appearance probability used here is highly reliable. Therefore, the sample is categorized as a usage time zone with a certain range of usage time, and the fraud appearance probability is calculated. However, since the category itself is not a real number, it cannot be used for score calculation as it is. Therefore, it is necessary to make the categories real numbers.

  Therefore, in order to sort the continuous variables into categories, assign real values to each category, and rearrange the assigned real values as meaningful variables, the fraud appearance probability corresponding to each category (p) Is converted into a logit value by applying to the equation of Equation 1.

  When the categories are converted into real numbers by logit values in this way, the realized categories are placed on the horizontal axis, and the fraudulent appearance probability is placed on the vertical axis, and FIG. 4 (use time zone is categorized every hour), FIG. As shown in the example of categorizing the use time zone in the range of 3 to 9 hours, it can be converted into a continuous variable that rises to the right. In this way, the relationship between the realized category and the fraud appearance probability is stored in the logit value conversion table shown in the example of FIG. 6 in the case of the categorization example of FIG.

  Further, from the authorization data, various items that can be used for calculating a score such as a usage amount in addition to the usage time are specified. The items used here are not limited to those included in the authorization data, but include items generated from the items included in the authorization data, items included in the history data, items included in the authorization data, and items included in the history data. The generated item may be the target. For item values corresponding to these items, a logit value conversion table can be generated in the same way as in the case of usage time, but when calculating scores using multiple items, each item category is The problem is how to weight the corresponding fraud appearance probability.

  Therefore, logistic regression analysis is performed on all items used for score calculation, and a regression coefficient corresponding to each item is calculated. The calculated regression coefficient is stored in a regression coefficient table as shown in the example of FIG. As described above, the logit value conversion table and the regression coefficient table created from the past authorization data and the like are used for score calculation for each authorization data.

  When the authorization data for score calculation is accepted, the item value of the item included in the authorization data used for score calculation, the item value of the item generated from the item included in the authorization data, the user specified from the authorization data The item value of the item included in the history data, the item value of the item generated from the item included in the authorization data and the item included in the history data are specified. With respect to these item values, a category is specified with reference to the logit value conversion table, respectively, and a logit value and an illegal appearance probability corresponding to the specified category are acquired.

  In addition, for the item for which the item value is specified, the regression coefficient corresponding to each item is acquired from the regression coefficient table. A value corresponding to “Zn” is calculated from the logit value, fraud appearance probability, and regression coefficient for each item obtained in this way, using the equation (2). The logit value average used here may be obtained by storing the logit value in each item in a part of the logit model and obtaining it. As long as the probability and the regression coefficient are reflected, subtraction of the logit average is not particularly limited.

  Further, for “Zn” calculated for each item, the total “Zsum” is calculated as shown in Equation 3.

  Finally, using the calculated “Zsum”, a score reflecting the possibility of fraud is calculated from the equation (4). The score calculation formula shown here is an example for calculating a highly accurate score, and scoring in the present invention is not limited to such calculation formula.

  In the score calculation in the scoring system according to the present invention, a method for specifying the logit value from items included in the authorization data will be described in detail below with reference to FIGS. It is assumed that the authorization data generated by the use of a credit card and subject to score calculation includes the items and item values shown in the example of FIG.

  When the authorization data in FIG. 8 is received, the behavior database is searched using the customer code included therein as a key, and the user's history data is acquired. FIG. 9 shows an example of the acquired history data, in which item values are recorded in items relating to the usage amount of the user at the previous use and items relating to the usage history such as the usage amount for the previous n times. ing.

  Next, from the authorization data in FIG. 8 and the history data in FIG. 9, data in which item values for score calculation shown in the example in FIG. 10 are specified is created. These items include items in which items and item values of authorization data and history data are used as they are, and items in which item values calculated from a plurality of items are provided with new items.

  In the example of FIG. 10, the item values of the item numbers c001 to c005 are all the item values of the items included in the authorization data. The host reception time (card usage time) is recorded as 12:00, and the unauthorized appearance probability of past authorization data in the time zone including this time is reflected in the score. A code indicating the transaction type is recorded in the transaction type, and a code indicating the payment type is recorded in the payment category. Similarly, the probability of unauthorized appearance of past authorization data in these codes is reflected in the score.

  In the example of FIG. 10, the item value of item number c011 is the item value of the item generated from the item value of the item included in the authorization data. The authorization data contains the usage limit set for the card and the available amount obtained by subtracting the total of the unpaid usage amount from the usage limit, but the usage reflects the extent to which the card is used. A value obtained by dividing the possible amount by the usage limit amount is set as a new item for score calculation, and the unauthorized appearance probability of past authorization data within the range including this value is reflected in the score.

  In the example of FIG. 10, the item values of the item numbers c021 to c027 are all item values of items generated from the item values of the items included in the authorization data and the item values of the items included in the history data. The ratio between the usage amount included in the authorization data and the previous usage amount included in the history data is recorded in the ratio to the previous usage amount, but the probability of unauthorized appearance of past authorization data in the range including this ratio is recorded. Is reflected in the score. For items such as match / mismatch with the previous product code, match / mismatch with the last destination country code, etc., items generated from the item value of the item included in the authorization data and the item value of the item included in the history data, respectively A code indicating coincidence / non-coincidence is recorded (for example, “1” is coincidence and “2” is non-coincidence), and the unauthorized appearance probability of past authorization data in each coincidence / non-coincidence is used as a score. reflect. The usage time score and usage day score are scores calculated by comparing the host reception date and reception date and time included in the authorization data with the usage frequency coefficient for each time zone and the usage frequency coefficient for each day of the history data. The unauthorized appearance probability of past authorization data in the range including the calculated score is reflected in the score.

  When data in which item values for score calculation are specified is created as in the example of FIG. 10, these item values are converted into logit values. The method and significance of conversion to logit values are as described above with reference to FIGS. 3 to 7. However, logit values in the range including item values for each item are obtained from the logit value conversion table. The logit value in each item is specified as in the example of FIG. The score is calculated by applying the arithmetic expressions of Equations 2 to 4 described above using the logit value specified in this way and the regression coefficient of each item acquired from the regression coefficient table.

  The flow of the scoring method according to the present invention will be described using the flowchart of FIG. First, authorization data for which a score is to be calculated is received from a host computer or the like of a credit card company (S1), and a customer code for specifying a user included in the authorization data (a card number or the like is used for specifying a user) (S2) may be identified.

  Subsequently, history data corresponding to the identified customer code is acquired from the database (S3), and items used for score calculation are selected or generated from the items included in the authorization data and the history data, and items corresponding to the respective items A value is specified (S4). Each identified item value is converted to a logit value corresponding to the range including the item value with reference to the logit value conversion table (S5).

  Furthermore, in order to weight the contribution to the score of each item, the regression coefficient corresponding to each item is obtained from the regression coefficient table (S6), and the regression coefficient corresponding to each converted logit value is applied (S6). S7). A score is calculated from the logit value and the regression coefficient based on a predetermined calculation formula (S8), and the calculated score is returned to the host computer of the credit card company (S9). Further, the history data is updated and stored in the database by reflecting the received authorization data (S10).

It is a figure which shows the outline | summary of the scoring system concerning this invention. It is a block diagram which shows the structure of the scoring system concerning this invention. It is a figure which shows an example of the fraud appearance probability according to utilization time slot | zone. It is a figure which shows the 1st example which converted and rearranged the irregular appearance probability according to use time zone into a logit value. It is a figure which shows the 2nd example which changed and rearranged the irregular appearance probability according to use time zone into a logit value. It is a figure which shows an example of the logit value conversion table in the scoring system concerning this invention. It is a figure which shows an example of the regression coefficient table in the scoring system concerning this invention. It is a figure which shows an example of the authorization data used in the scoring system concerning this invention. It is a figure which shows an example of the historical data used in the scoring system concerning this invention. It is a figure which shows an example of the data by which the item value for the score calculation used in the scoring system concerning this invention was specified. It is a figure which shows an example of the data by which the item value for the score calculation used in the scoring system concerning this invention was converted into the logit value. It is a flowchart which shows the flow of the scoring method concerning this invention.

Explanation of symbols

10 Scoring system 11 Data transmission / reception unit 12 Authorization DB
13 Model analysis unit 131 Logit value conversion analysis unit 132 Logistic regression analysis unit 133 Logit value conversion table 134 Regression coefficient table 14 Scoring unit 141 Authorization data reception unit 142 Behavior data acquisition unit 143 Score calculation unit 144 Behavior update unit 15 Behavior DB
151 Viavia item creation unit 20 Card company host computer 30 CAT terminal

Claims (8)

A scoring system for calculating a score that reflects the probability of unauthorized use of a credit card,
Authorization data storage means for storing authorization data received every time a credit card is used;
History data storage means for storing, for each user, history data relating to past credit card usage history of the credit card user;
Per predetermined item included in the authorization data stored in the authorization Item value categories included in the authorization data stored in the data storage means, and said authorization data storing means, items of the items included in the authorization data values and, for the history category of data storage item value of the new item indicating a change in the item value of the item included in the history data stored in the unit, the occurrence of fraud ratio fraud by category in each item appeared A logit value conversion table for storing the calculated logit value in association with each category for each item;
For the item included in the authorization data stored in the authorization data storage means and the predetermined item included in the authorization data stored in the authorization data storage means, the item value of the item included in the authorization data, and the history for a new item indicating a change in the item value of the item included in the stored historical data in the data storage means, a contribution to the score of each item were calculated by logistic regression analysis, in association with each of the item storage Regression coefficient table,
An authorization data receiving means for receiving authorization data relating to a credit card usage request;
Item value specifying means for specifying an item value of a predetermined item included in the authorization data received by the authorization data receiving means;
First category specifying means for specifying a category corresponding to the item value specified by the item value specifying means;
History data acquisition means for reading and acquiring history data corresponding to a user specified from the authorization data received by the authorization data receiving means, from the history data storage means;
For a given item that the authorization data receiving unit is included in the authorization data received, the change in the item value of the item included in the authorization data, item values of the items included in the history data which the history data acquisition unit acquires Item value calculation means for calculating the item value of a new item indicating
Second category specifying means for specifying a category corresponding to the item value calculated by the item value calculating means ;
Referring to the logit value conversion table, first logit value acquisition means for reading out and acquiring logit values corresponding to the category specified by the first category specifying means from the logit value conversion table;
With reference to the regression coefficient table, a first regression coefficient acquisition unit that reads and acquires a regression coefficient corresponding to the item for which the first category identification unit has identified a category from the regression coefficient table;
Referring to the logit value conversion table, second logit value acquisition means for reading out and acquiring the logit value corresponding to the category specified by the second category specifying means from the logit value conversion table;
Referring to the regression coefficient table, second regression coefficient acquisition means for reading out and acquiring the regression coefficient corresponding to the item for which the second category identification means has identified the category from the regression coefficient table;
For each item corresponding to the logit value , the logit value obtained by reading from the logit value conversion table including the logit value acquired by the first logit value acquiring unit and the logit value acquired by the second logit value acquiring unit , The value obtained by multiplying the regression coefficient obtained by reading from the regression coefficient table including the regression coefficient acquired by the first regression coefficient acquisition unit and the second regression coefficient acquisition unit is added, and the credit card Score calculating means for calculating a score reflecting the probability that the usage request is invalid,
A scoring system comprising: With respect to the authorization data received by the authorization data receiving means, an item value of a predetermined item included in the history data stored in the history data storage means corresponding to a credit card user of the authorization data is obtained as the authorization data. The scoring system according to claim 1, further comprising history data updating means for updating the item value of the item included in the item . In the logit value conversion table, item value categories of new items calculated from item values of two or more items included in the authorization data stored in the authorization data storage means are stored in the item values of each item. The logit value calculated from the fraud appearance ratio where fraud appeared by category is stored in association with each category for each item,
In the regression coefficient table, the new item for calculating the item value from the item values of two or more items included in the authorization data stored in the authorization data storage means is calculated for each item calculated by logistic regression analysis. The contribution to the score is stored in association with each item,
Second item value calculating means for calculating item values of the new item from item values of two or more items included in the authorization data for a predetermined item included in the authorization data received by the authorization data receiving means; ,
Third category specifying means for specifying a category corresponding to the item value calculated by the second item value calculating means;
Referring to the logit value conversion table, third logit value acquisition means for reading out and acquiring the logit value corresponding to the category specified by the third category specifying means from the logit value conversion table;
A third regression coefficient acquisition unit that reads the regression coefficient corresponding to the item for which the third category identification unit has identified the category from the regression coefficient table with reference to the regression coefficient table; And
The score calculation means uses a value obtained by multiplying the logit value acquired by the third logit value acquisition means by the regression coefficient acquired by the third regression coefficient acquisition means for an item corresponding to the logit value. The scoring system according to claim 1 or 2, wherein the score is calculated . In the logit value conversion table, item value categories of new items calculated from item values of two or more items included in the history data stored in the history data storage means are stored in the item values of each item. The logit value calculated from the fraud appearance ratio where fraud appeared by category is stored in association with each category for each item,
In the regression coefficient table, the new item for calculating the item value from the item values of two or more items included in the history data stored in the history data storage means is calculated for each item calculated by logistic regression analysis. The contribution to the score is stored in association with each item,
Third item value calculating means for calculating an item value of the new item from item values of two or more items included in the history data acquired by the history data acquiring means;
Fourth category specifying means for specifying a category corresponding to the item value calculated by the third item value calculating means;
Referring to the logit value conversion table, fourth logit value acquisition means for reading out and acquiring logit values corresponding to each category specified by the fourth category specifying means from the logit value conversion table;
Referring to the regression coefficient table, fourth regression coefficient acquisition means for reading out and acquiring a regression coefficient corresponding to each item for which the fourth category specification means has specified a category from the regression coefficient table; Have
The score calculation means is obtained by multiplying the logit value acquired by the fourth logit value acquisition means by the regression coefficient acquired by the fourth regression coefficient acquisition means for each item corresponding to the logit value. The scoring system according to claim 1 or 2, wherein the score is calculated by using the scoring system. A scoring method for calculating a score that reflects the probability of unauthorized use of a credit card,
The computer system for calculating the score uses an authorization data storage unit that stores authorization data received every time a credit card is used, and history data regarding past credit card usage history of the credit card user. A history data storage section for storing each person,
It said computer system, per predetermined item included the authorization Item value categories included in the authorization data stored in the data storage unit, and the authorization data stored in the authorization data storage unit, included in the authorization data and item value of the item, for the history category of item value of the new item indicating a change in the item value of the item included in the stored historical data in the data storage unit, it is invalid by category in each of the items appearing to be Storing the logit value calculated from the fraudulent appearance ratio in the logit value conversion table in association with each category for each item;
Said computer system, said authorization items included in authorization data stored in the data storage unit, and per predetermined item included in the authorization data stored in the authorization data storage unit, items of the items included in the authorization data and values for the new item indicating a change in the item value of the item included in the history data stored in the history data storage unit, the contribution to the score of each item were calculated by logistic regression analysis, each Storing in the regression coefficient table in association with the item;
The logit value conversion table and regression coefficient table data are updated,
When calculating a score that reflects the probability of fraudulent credit card usage,
An authorization data receiving step in which the computer system receives authorization data relating to a credit card usage request;
An item value specifying step for specifying an item value of a predetermined item included in the authorization data received in the authorization data receiving step by the computer system;
A first category specifying step in which the computer system specifies a category corresponding to the item value specified in the item value specifying step;
A history data acquisition step in which the computer system reads and acquires history data corresponding to a user specified from the authorization data received in the authorization data reception step , from the history data storage unit;
The items the computer system, for a given item included in the authorization data received by the authorization data receiving step, included the item value of the item included in the authorization data, the history data acquired by the history data acquiring step An item value calculation step for calculating an item value of a new item indicating a change in the item value of
A second category specifying step in which the computer system specifies a category corresponding to the item value calculated in the item value calculating step;
A first logit value obtaining step in which the computer system reads and obtains a logit value corresponding to the category specified in the first category specifying step from the logit value conversion table with reference to the logit value conversion table; ,
A first regression coefficient acquisition step in which the computer system reads and acquires a regression coefficient corresponding to the item whose category is specified in the first category specification step with reference to the regression coefficient table; ,
A second logit value acquisition step in which the computer system reads the logit value corresponding to the category specified in the second category specifying step from the logit value conversion table with reference to the logit value conversion table. When,
A second regression coefficient acquisition step in which the computer system reads and acquires a regression coefficient corresponding to the item whose category is specified in the second category specification step with reference to the regression coefficient table. When,
Each logit value obtained by reading from the logit value conversion table including the logit value acquired by the computer system in the first logit value acquisition step and the second logit value acquisition step corresponds to the logit value. For the item, the values obtained by adding the regression coefficients obtained by reading from the regression coefficient table including the regression coefficients acquired in the first regression coefficient acquisition step and the second regression coefficient acquisition step are added together, and the added value A score calculating step for calculating a score reflecting the probability that the credit card use request is fraudulent,
A scoring method comprising: The item value of a predetermined item included in the history data stored in the history data storage unit corresponding to the credit card user of the authorization data for the authorization data received in the authorization data receiving step by the computer system The scoring method according to claim 5 , further comprising: a history data update step of updating the value to an item value of the item included in the authorization data . In the logit value conversion table, item value categories of new items calculated from item values of two or more items included in the authorization data stored in the authorization data storage unit are stored in the item values of each item. The logit value calculated from the fraud appearance ratio where fraud appeared by category is stored in association with each category for each item,
In the regression coefficient table, the new item for calculating the item value from the item values of two or more items included in the authorization data stored in the authorization data storage unit is calculated for each item calculated by logistic regression analysis. The contribution to the score is stored in association with each item,
The computer system calculates, for a predetermined item included in the authorization data received in the authorization data receiving step, an item value of the new item from item values of two or more items included in the authorization data. Item value calculation step,
A third category specifying step in which the computer system specifies a category corresponding to the item value calculated in the second item value calculating step;
A third logit value acquisition step in which the computer system reads and acquires the logit value corresponding to the category specified in the third category specifying step with reference to the logit value conversion table. When,
A third regression coefficient acquisition step in which the computer system reads and acquires a regression coefficient corresponding to the item whose category is specified in the third category identification step with reference to the regression coefficient table. And having
In the score calculation step, the computer system multiplies the logit value acquired in the third logit value acquisition step by the regression coefficient acquired in the third regression coefficient acquisition step for an item corresponding to the logit value. The scoring method according to claim 5 or 6, characterized in that the score is calculated using a measured value . In the logit value conversion table, item value categories of new items calculated from item values of two or more items included in the history data stored in the history data storage unit are stored in the item values of each item. The logit value calculated from the fraud appearance ratio where fraud appeared by category is stored in association with each category for each item,
In the regression coefficient table, new items for calculating item values from item values of two or more items included in the history data stored in the history data storage unit are calculated for each item calculated by logistic regression analysis. The contribution to the score is stored in association with each item,
A third item value calculating step in which the computer system calculates an item value of the new item from item values of two or more items included in the history data acquired in the history data acquiring step;
A fourth category specifying step in which the computer system specifies a category corresponding to the item value calculated in the third item value calculating step;
A fourth logit value acquisition step in which the computer system reads out and acquires a logit value corresponding to the category specified in the fourth category specifying step with reference to the logit value conversion table from the logit value conversion table. When,
A fourth regression coefficient acquisition step in which the computer system reads and acquires a regression coefficient corresponding to the item whose category is specified in the fourth category identification step with reference to the regression coefficient table from the regression coefficient table. And having
In the score calculation step, the computer system multiplies the logit value acquired in the fourth logit value acquisition step by the regression coefficient acquired in the fourth regression coefficient acquisition step for the item corresponding to the logit value. The scoring method according to claim 5 or 6, characterized in that the score is calculated using a measured value .