Systems and methods for outlier detection of transactions

Abstract

Described are systems and methods for outlier detection and transaction monitoring. This may include collecting corporate data, determining features, detecting relevant features, determining a system model, determining a control policy, monitoring incoming transactions, determining if an outlier alert should be sent, transmitting the outlier alert, receiving user adjudication of the fraud alerts, and feedback of the adjudication. The systems and methods may use aspects of fuzzy logic, predictive modeling, network and community detection, outlier detection, and fuzzy aggregation.

We claim: 
     
         1 . A method for generating an outlier transaction identification model and a selected control policy within an enterprise network comprising a plurality of transaction processing sites and a plurality of enterprise servers, the method comprising:
 receiving, at a first server of the plurality of enterprise servers, transaction data from the plurality of transaction processing sites, the transaction data comprising at least one selected from the group of an insurance claim, a financial institution transaction, and an insurance claim disposition;   determining, at the first server, transformed transaction data based on the transaction data,   determining one or more features from the transformed transaction data;   determining one or more actionable features from the one or more features;   generating an outlier transaction identification model from the one or more actionable features; and   selecting a selected control policy for the outlier transaction identification model, wherein the outlier transaction identification model and the selected control policy cooperate with an intelligent agent to determine an outlier transaction identification alert.   
     
     
         2 . The method of  claim 1 , wherein the generating the outlier transaction identification model further comprises:
 determining an interaction l_jk^Pr comprising a j x k matrix, each element of the j x k matrix comprising a correlation between a revenue for product j and a fraud detection activity k based on the transformed transaction data;   determining an interaction l_jk^(O/H) comprising a M x P matrix, each element of the M x P matrix comprising a correlation between an overhead cost for a product M and a fraud detection activity P based on the transformed transaction data; and wherein the outlier transaction identification model further comprises the interaction l_jk^Pr and the interaction l_jk^(O/H).   
     
     
         3 . The method of  claim 2 , wherein the selecting the selected control policy further comprises:
 determining a coefficient C_p based on the transformed transaction data;   determining a coefficient β_(c(j)) based on the transformed transaction data; and wherein the selected control policy further comprises the coefficient C_p and the coefficient β_(c(j)).   
     
     
         4 . The method of  claim 3 , wherein the determining, at the intelligent agent, the coefficient C_p further comprises performing a gradient descent, and the determining, at the intelligent agent, a coefficient β_(c(j)) further comprises performing a gradient descent. 
     
     
         5 . The method of  claim 3 , wherein the determining, at the intelligent agent, the coefficient C_p further comprises performing a gradient descent, and the determining, at the intelligent agent, a coefficient β_(c(j)) further comprises performing an iterative optimization algorithm. 
     
     
         6 . The method of  claim 1 , further comprising performing a signal analysis to generate signal analysis data, wherein the transformed transaction data comprises the transaction data and the signal analysis data. 
     
     
         7 . The method of  claim 6 , wherein the signal analysis data comprises a power spectrum. 
     
     
         8 . The method of  claim 1 , wherein the feature determination comprises performing at least one selected from the group of a linear correlation, a principal components analysis, and least absolute shrinkage and selection operator (LASSO) regularized regression. 
     
     
         9 . The method of  claim 1 , wherein the outlier transaction identification model and the selected control policy are provided to an operational computer system using an application programming interface (API). 
     
     
         10 . A system for generating an outlier transaction identification model and a selected control policy within an enterprise network comprising a plurality of transaction processing sites and a plurality of enterprise servers, the system comprising:
 a first server in the plurality of enterprise servers, the first server comprising a memory and a processor in communication with the memory, the processor configured to: 
 receive transaction data from the plurality of transaction processing sites, the transaction data comprising at least one selected from the group of an insurance claim, 
 a financial institution transaction, and an insurance claim disposition; 
 determine transformed transaction data based on the transaction data, 
 determine one or more features from the transformed transaction data; 
 determine one or more actionable features from the one or more features; 
 generate an outlier transaction identification model from the one or more actionable features; and 
 select a selected control policy for the outlier transaction identification model, 
 wherein the outlier transaction identification model and the selected control policy cooperate with an intelligent agent to determine an outlier transaction identification alert. 
   
     
     
         11 . The system of  claim 10  wherein the processor is further configured to generate the outlier transaction identification model by:
 determining an interaction l_jk^Pr comprising a j x k matrix, each element of the j x k matrix comprising a correlation between a revenue for product j and a fraud detection activity k based on the transformed transaction data; 
 determining an interaction l_jk^(O/H) comprising a M x P matrix, each element of the M x P matrix comprising a correlation between an overhead cost for a product M and a fraud detection activity P based on the transformed transaction data; and 
 wherein the outlier transaction identification model further comprises the interaction l_jk^Pr and the interaction l_jk^(O/H). 
 
     
     
         12 . The system of  claim 11  wherein the processor is further configured to select the selected control policy further by:
 determining a coefficient C_p based on the transformed transaction data; 
 determining a coefficient β_(c(j)) based on the transformed transaction data; and wherein the selected control policy further comprises the coefficient C_p and the coefficient β_(c(j)). 
 
     
     
         13 . The system of  claim 12  wherein the processor is further configured to determine the coefficient C_p by performing a gradient descent, and the determining, at the intelligent agent, a coefficient β_(c(j)) further comprises performing a gradient descent. 
     
     
         14 . The system of  claim 12 , wherein the processor is further configured to determine the coefficient C_p by performing a gradient descent, and the determining, at the intelligent agent, a coefficient β_(c(j)) further comprises performing an iterative optimization algorithm. 
     
     
         15 . The system of  claim 10 , wherein the processor is further configured to perform a signal analysis to generate signal analysis data, wherein the transformed transaction data comprises the transaction data and the signal analysis data. 
     
     
         16 . The system of  claim 15 , wherein the signal analysis data comprises a power spectrum. 
     
     
         17 . The system of  claim 10 , wherein the processor is further configured to perform feature determination by performing at least one selected from the group of a linear correlation, a principal components analysis, and a least absolute shrinkage and selection operator (LASSO) regularized regression. 
     
     
         18 . The system of  claim 10 , wherein the processor is further configured to provide the outlier transaction identification model and the selected control policy to an operational computer system using an application programming interface (API). 
     
     
         19 . A computer program product comprising computer-readable instructions carried on a computer readable medium which, when executed by a processor, cause the processor to perform a method for generating an outlier transaction identification model and a selected control policy within an enterprise network comprising a plurality of transaction processing sites and a plurality of enterprise servers, the method comprising:
 receiving, at a first server of the plurality of enterprise servers, transaction data from the plurality of transaction processing sites, the transaction data comprising at least one selected from the group of an insurance claim, a financial institution transaction, and an insurance claim disposition;   determining, at the first server, transformed transaction data based on the transaction data,   determining one or more features from the transformed transaction data;   determining one or more actionable features from the one or more features;   generating an outlier transaction identification model from the one or more actionable features; and   selecting a selected control policy for the outlier transaction identification model, wherein the outlier transaction identification model and the selected control policy cooperate with an intelligent agent to determine an outlier transaction identification alert.   
     
     
         20 . The computer program product of  claim 19 , wherein the generating the outlier transaction identification model further comprises:
 determining an interaction l_jk^Pr comprising a j x k matrix, each element of the j x k matrix comprising a correlation between a revenue for product j and a fraud detection activity k based on the transformed transaction data;   determining an interaction l_jk^(O/H) comprising a M x P matrix, each element of the M x P matrix comprising a correlation between an overhead cost for a product M and a fraud detection activity P based on the transformed transaction data; and wherein the outlier transaction identification model further comprises the interaction l_jk^Pr and the interaction l_jk^(O/H).