Technique to aggregate merchant level information for use in a supervised learning model to detect recurring trends in consumer transactions

Abstract

A method is disclosed, comprising: aggregating a plurality of sets of transactions, each set of transactions comprising transactions related to an account-merchant pairing; determining variables characterizing a recurrence period based on transaction dates of the transactions in the each set; predicting the recurrence period for a transaction related to the account-merchant pairing for each customer of the plurality of customers; aggregating another set of transactions between the plurality of customers and the merchant; evaluating a distribution of the recurrence period for each customer within range of a distant point; and based on the evaluation of the distribution of the recurrence period for each customer, generating a probability of the merchant having a recurrent transaction with the customer. The account-merchant pairing may comprise a customer account and a merchant of a plurality of customers and merchants. Another set of transactions may comprise transactions in the plurality of sets of transactions.

What is claimed is: 
     
         1 . A method, comprising:
 forming a set of transactions from a plurality of transactions related to an account-merchant pairing, wherein the set of transactions spans a particular period of time;   identifying a split date within the particular period of time;   generating, based on the split date, an analysis portion comprising a first subset of transactions of the set of transactions and holdout portion comprising a second subset of transactions of the set of transactions;   determining a vector strength, a coverage, and a redundancy, wherein the vector strength, the coverage, and the redundancy are phase variables characterizing a recurrence period within the analysis portion;   generating, based on the recurrence period, a predicted future transaction date;   identifying one or more matching transactions in the holdout portion based on the predicted future transaction date;   generating a target label for one or more transactions in the analysis portion responsive to identifying the one or more matching transactions;   training a machine learning model by using the target label;   identifying a second account-merchant pairing associated with a new transaction; and   generating, by the trained machine learning model, a probability of a recurring transaction within a set of transactions associated with the second account-merchant pairing.   
     
     
         2 . The method of  claim 1 , wherein the first subset of transactions include transactions of the set of transactions that occurred prior to the split date and the second subset of transactions include transactions of the set of transactions that occurred after the split date. 
     
     
         3 . The method of  claim 1 , wherein identifying the one or more matching transactions further comprises:
 identifying a first potential matching transaction in the holdout portion closest in time to the predicted future transaction date;   identifying a transaction date of the first potential matching transaction;   calculating a difference between the transaction date of the first potential matching transaction and the predicted future transaction date; and   comparing the difference to a date tolerance parameter that identifies a maximum allowed difference to the predicted future transaction date.   
     
     
         4 . The method of  claim 1 , wherein a number of the one or more matching transactions is determined based on a predictions parameter that identifies a number of matching transactions that are to be identified in the holdout portion. 
     
     
         5 . The method of  claim 1 , further comprising:
 determining a standard deviation of differences between transaction dates of transactions in the plurality of transactions;   determining an average difference between the transaction dates of the transactions in the plurality of transactions; and   determining a ratio between the standard deviation of difference between the transaction dates and the average difference between the transaction dates, wherein the standard deviation of the transaction dates, the average difference between the transaction dates, and the ratio of the standard deviation of the transaction dates to the average difference between the transaction dates are delta time variables further characterizing the recurrence period.   
     
     
         6 . The method of  claim 1 , further comprising:
 converting the transactions dates of the transactions in the plurality of transactions into ordinal dates;   mapping the ordinal dates into a phase space, wherein the phase space represents a billing cycle;   determining a mean coordinate based on the mapped ordinal dates;   determining the vector strength as a distance between (0,0) coordinate of a unit circle and the mean coordinate;   determining the coverage, wherein the coverage is a count of billing cycles in the phase space having at least one transaction; and   determining the redundancy, wherein the redundancy is a percentage of billing cycles having more than one transaction.   
     
     
         7 . The method of  claim 1 , the method further comprising determining a modified vector strength as 
       
         
           
             
               
                 
                   r 
                   adjusted 
                 
                 = 
                 
                   1 
                   - 
                   
                     
                       2 
                       π 
                     
                     ⁢ 
                     
                       arccos 
                       ⁡ 
                       
                         ( 
                         r 
                         ) 
                       
                     
                   
                 
               
               , 
             
           
         
       
       wherein r adjusted  is the modified vector strength and r is the vector strength. 
     
     
         8 . The method of  claim 1 , further comprising determining a probability of the transactions in the set of transactions based on a Euclidean distance between an ideal point and a point representing the vector strength, the coverage, and the redundancy, wherein the ideal point represents the vector strength of value 1, the coverage of value 1, and the redundancy of value 0 as (1,1,0), and wherein each distance between the phase variable coordinates corresponds to one of a weekly, a biweekly, a monthly, a bimonthly, a quarterly, a semi-annually, and a yearly period. 
     
     
         9 . A system, comprising:
 a memory for storing instructions; and   a processor, communicatively coupled to the memory, configured to execute the instructions, the instructions causing the processor to:
 form a set of transactions from a plurality of transactions related to an account-merchant pairing, wherein the set of transactions spans a particular period of time; 
 identify a split date within the particular period of time; 
 generate, based on the split date, an analysis portion comprising a first subset of transactions of the set of transactions and holdout portion comprising a second subset of transactions of the set of transactions; 
 determine a vector strength, a coverage, and a redundancy, wherein the vector strength, the coverage, and the redundancy are phase variables characterizing a recurrence period within the analysis portion; 
 generate, based on the recurrence period, a predicted future transaction date; 
 identify one or more matching transactions in the holdout portion based on the predicted future transaction date; 
 generate a target label for one or more transactions in the analysis portion responsive to identifying the one or more matching transactions; 
 train a machine learning model by using the target label; 
 identify a second account-merchant pairing associated with a new transaction; and 
 generate, by the trained machine learning model, a probability of a recurring transaction within a set of transactions associated with the second account-merchant pairing. 
   
     
     
         10 . The system of  claim 9 , wherein the first subset of transactions include transactions of the set of transactions that occurred prior to the split date and the second subset of transactions include transactions of the set of transactions that occurred after the split date. 
     
     
         11 . The system of  claim 9 , wherein to identify the one or more matching transactions, the processor is further configured to:
 identify a first potential matching transaction in the holdout portion closest in time to the predicted future transaction date;   identify a transaction date of the first potential matching transaction;   calculate a difference between the transaction date of the first potential matching transaction and the predicted future transaction date; and   compare the difference to a date tolerance parameter that identifies a maximum allowed difference to the predicted future transaction date.   
     
     
         12 . The system of  claim 9 , wherein a number of the one or more matching transactions is determined based on a predictions parameter that identifies a number of matching transactions that are to be identified in the holdout portion. 
     
     
         13 . The system of  claim 9 , wherein to determine the variables characterizing the recurrence period, the processor is further configured to:
 determine a standard deviation of differences between transaction dates of transactions in the plurality of transactions;   determine an average difference between the transaction dates of the transactions in the plurality of transactions; and   determine a ratio between the standard deviation of difference between the transaction dates and the average difference between the transaction dates, wherein the standard deviation of the transaction dates, the average difference between the transaction dates, and the ratio of the standard deviation of the transaction dates to the average difference between the transaction dates are delta time variables further characterizing the recurrence period.   
     
     
         14 . The system of  claim 9 , the processor is further configured to:
 convert the transactions dates of the transactions in the plurality of transactions into ordinal dates;   map the ordinal dates into a phase space, wherein the phase space represents a billing cycle;   determine a mean coordinate based on the mapped ordinal dates;   determine the vector strength as a distance between (0,0) coordinate of a unit circle and the mean coordinate;   determine the coverage, wherein the coverage is a count of billing cycles in the phase space having at least one transaction; and   determine the redundancy, wherein the redundancy is a percentage of billing cycles having more than one transaction.   
     
     
         15 . The system of  claim 1 , the processor is further configured to:
 determine a modified vector strength as   
       
         
           
             
               
                 
                   r 
                   adjusted 
                 
                 = 
                 
                   1 
                   - 
                   
                     
                       2 
                       π 
                     
                     ⁢ 
                     
                       arccos 
                       ⁡ 
                       
                         ( 
                         r 
                         ) 
                       
                     
                   
                 
               
               , 
             
           
         
       
       wherein r adjusted  is the modified vector strength and r is the vector strength. 
     
     
         16 . The system of  claim 13 , wherein the processor is further configured to:
 determine a probability of the transactions in the set of transactions based on a Euclidean distance between an ideal point and a point representing the vector strength, the coverage, and the redundancy, wherein the ideal point represents the vector strength of value 1, the coverage of value 1, and the redundancy of value 0 as (1,1,0), and wherein each distance between the phase variable coordinates corresponds to one of a weekly, a biweekly, a monthly, a bimonthly, a quarterly, a semi-annually, and a yearly period.   
     
     
         17 . A non-transitory, tangible computer-readable device having instructions stored thereon that, when executed by at least one computing device, causes the at least one computing device to perform operations comprising:
 forming a set of transactions from a plurality of transactions related to an account-merchant pairing, wherein the set of transactions spans a particular period of time;   identifying a split date within the particular period of time;   generating, based on the split date, an analysis portion comprising a first subset of transactions of the set of transactions and holdout portion comprising a second subset of transactions of the set of transactions;   determining a vector strength, a coverage, and a redundancy, wherein the vector strength, the coverage, and the redundancy are phase variables characterizing a recurrence period within the analysis portion;   generating, based on the recurrence period, a predicted future transaction date;   identifying one or more matching transactions in the holdout portion based on the predicted future transaction date;   generating a target label for one or more transactions in the analysis portion responsive to identifying the one or more matching transactions;   training a machine learning model by using the target label;   identifying a second account-merchant pairing associated with a new transaction; and   generating, by the trained machine learning model, a probability of a recurring transaction within a set of transactions associated with the second account-merchant pairing.   
     
     
         18 . The non-transitory, tangible computer-readable device of  claim 17 , wherein the first subset of transactions include transactions of the set of transactions that occurred prior to the split date and the second subset of transactions include transactions of the set of transactions that occurred after the split date. 
     
     
         19 . The non-transitory, tangible computer-readable device of  claim 17 , wherein to identify the one or more matching transactions, the operations further comprise:
 identifying a first potential matching transaction in the holdout portion closest in time to the predicted future transaction date;   identifying a transaction date of the first potential matching transaction;   calculating a difference between the transaction date of the first potential matching transaction and the predicted future transaction date; and   comparing the difference to a date tolerance parameter that identifies a maximum allowed difference to the predicted future transaction date   
     
     
         20 . The non-transitory, tangible computer-readable device of  claim 17 , wherein a number of the one or more matching transactions is determined based on a predictions parameter that identifies a number of matching transactions that are to be identified in the holdout portion