Adaptive Fraud Detection

Abstract

A computer-implemented method for technologically improving a computer-implemented machine-learning model, the method comprising receiving, by a model, at least a first data record; generating a first score representing a first likelihood that the first data record is associated with a first classification, in response to feedback received from one or more data sources communicating with at least one computing system on which the model is implemented; generating a second score to represent a second likelihood that the first data record is associated with the first classification, in response to the first score being higher than a threshold value.

What is claimed is: 
     
         1 . A method for technologically improving a computer-implemented machine-learning model, the method comprising:
 receiving, by a model, at least a first data record;   generating a first score representing a first likelihood that the first data record is associated with a first classification, in response to feedback received from one or more data sources communicating with at least one computing system on which the model is implemented;   generating a second score to represent a second likelihood that the first data record is associated with the first classification, in response to the first score being higher than a threshold value,   the feedback comprising information about at least a second data record received by the model prior to the model receiving the first data record, the second data record being identified as associated with at least one of the first classification or a second classification according to the feedback; and   applying at least one of the first score, the second score, or a blend of the first score and the second score to indicate likelihood of occurrence of an event associated with the first classification.   
     
     
         2 . The method of  claim 1 , wherein one or more scoring parameters are used by the model to generate the second score, the feedback being used adaptively to update the one or more scoring parameters. 
     
     
         3 . The method of  claim 1 , wherein a blended score determined based on the first score and the second score is determined and displayed on a graphical user interface to alert a human operator of a likelihood of occurrence of the undesirable event. 
     
     
         4 . The method of  claim 2 , wherein the model receives the information about the second data record as input variables to update the one or more scoring parameters used by the model to generate the second score. 
     
     
         5 . The method of  claim 1 , wherein at least one of the one or more data sources stores the second record in a first-in first-out (FIFO) storage data structure. 
     
     
         6 . The method of  claim 1 , further comprising:
 computing probabilities of the second data record being associated with the first classification;   comparing the first data record with at least the second data record; and   computing the second likelihood that the first data record is associated with the first classification based on results of the comparing of the first data record with at least the second data record.   
     
     
         7 . The method of  claim 6 , further comprising combining the second likelihood with the probabilities of at least the second data record being associated with the first classification to calculate marginal probabilities of the first data record being associated with the first classification. 
     
     
         8 . The method of  claim 7 , further comprising combining the marginal probabilities to compute a posterior probability of the first data record. 
     
     
         9 . The method of  claim 8 , wherein the second score is based at least in part on the posterior probability. 
     
     
         10 . The method of  claim 1 , wherein in response to determining whether the event is associated with the first classification, feeding corresponding records and associated feature variables to the model to adaptively enhance the model's classification accuracy. 
     
     
         11 . A computer program product comprising machine-readable media having computer program code that is configured to instruct a programmable processor to:
 receive, by a model, at least a first data record;   generate a first score representing a first likelihood that the first data record is associated with a first classification, in response to feedback received from one or more data sources communicating with at least one computing system on which the model is implemented;   generate a second score to represent a second likelihood that the first data record is associated with the first classification, in response to the first score being higher than a threshold value,   the feedback comprising information about at least a second data record received by the model prior to the model receiving the first data record, the second data record being identified as associated with at least one of the first classification or a second classification according to the feedback; and   apply at least one of the first score, the second score, or a blend of the first score and the second score to indicate likelihood of occurrence of an event associated with the first classification.   
     
     
         12 . The computer program product of  claim 11 , wherein one or more scoring parameters are used by the model to generate the second score, the feedback used to update the one or more scoring parameters. 
     
     
         13 . The computer program product of  claim 11 , wherein a blended score determined based on the first score and the second score is determined and displayed on a graphical user interface to alert a human operator of likelihood of occurrence of the undesirable event. 
     
     
         14 . The computer program product of  claim 12 , wherein the model adaptively receives the information about the second data record as input variables provided to update the one or more scoring parameters used by the model to generate the second score. 
     
     
         15 . The computer program product of  claim 11 , wherein at least one of the one or more data sources stores the second record in a first-in first-out (FIFO) storage data structure. 
     
     
         16 . The computer program product of  claim 11 , where:
 probabilities of the second data record being associated with the first classification are computed;   the first data record is compared with at least the second data record; and   the second likelihood that the first data record is associated with the first classification is computed based on the comparing the first data record with at least the second data record.   
     
     
         17 . A computer-implemented system comprising:
 at least one programmable processor; and   a non-transitory machine-readable medium storing instructions that, when executed by the at least one programmable processor, cause the at least one programmable processor to perform operations comprising:   receiving, by a model, at least a first data record;   generating a first score representing a first likelihood that the first data record is associated with a first classification, in response to feedback received from one or more data sources communicating with at least one computing system on which the model is implemented;   generating a second score to represent a second likelihood that the first data record is associated with the first classification, in response to the first score being higher than a threshold value,   the feedback comprising information about at least a second data record received by the model prior to the model receiving the first data record, the second data record being identified as associated with at least one of the first classification or a second classification according to the feedback; and   applying at least one of the first score, the second score, or a blend of the first score and the second score to indicate likelihood of occurrence of an event associated with the first classification.   
     
     
         18 . The system of  claim 17 , wherein one or more scoring parameters are used by the model to generate the second score, the feedback used to update the one or more scoring parameters. 
     
     
         19 . The system of  claim 17 , wherein the model adaptively receives the information about the second data record as input variables provided to update the one or more scoring parameters used by the model to generate the second score. 
     
     
         20 . The system of  claim 17 , wherein an event associated with the first classification is identified as a fraudulent event, and wherein additional segmentation of fraud in the model by identifying various fraud types improves the model's performance.