Decision engine system and method

Abstract

System, apparatus, user equipment, and associated computer program and computing methods are provided for facilitating efficient decision-making with respect to a subject entity. In one aspect, a labeled training dataset containing N records respectively corresponding to N entities is provided for training a decision engine based on performing supervised learning. Responsive to receiving a plurality of attribute values for the subject entity requiring a decision relative to an estimate of a performance variable based on at least a portion of the attribute values, the trained decision engine is configured to determine a credit risk score as a function obtained as a set of linearly decomposed constituent components corresponding to the attribute values of the subject entity, thereby effectuating an objective determination of which attributes contribute to what portions of the score in a computationally efficient manner.

What is claimed is: 
     
         1 . A computer-implemented method, comprising:
 configuring a Gradient Boosted Tree (GBT) ensemble comprising a plurality of regression trees as a trained decision engine based on a training dataset containing N records corresponding to N entities respectively, each record comprising a value relating to a performance variable and a plurality of values corresponding to a set of attribute variables;   receiving, over a communications network, input values corresponding to a plurality of attributes associated with a subject entity;   determining, by the trained decision engine, a credit risk score for the subject entity responsive to the input values;   decomposing the credit risk score into a set of linearly decomposed constituent components corresponding to the attributes of the subject entity;   identifying a subset of the attributes of the subject entity contributing most to the credit risk score; and   performing a decision-making action with respect to the subject entity based at least in part upon one or more respective portions of the credit risk score associated with the subset of attributes of the subject entity.   
     
     
         2 . The computer-implemented method as recited in  claim 1 , further comprising training the GBT ensemble, the training including:
 constructing an initial regression tree based on minimization of a cost function associated with a root node corresponding to a select attribute variable and iteratively branching a training population into sub-populations corresponding to a set of leaf nodes based on remaining attribute variables;   determining a residual value and a corresponding loss function associated with the initial regression tree;   constructing a next regression tree based on the residual value and the loss function of the initial regression tree; and   iteratively generating subsequent regression trees based on a predecessor regression tree's residual value and corresponding loss function associated therewith, the iterative generation continuing until a predetermined number of regression trees are obtained as a fitted ensemble of the plurality of regression trees operable as the trained decision engine.   
     
     
         3 . The computer-implemented method as recited in  claim 2 , further comprising generating an action report regardless of an adverse decision based on the credit risk score. 
     
     
         4 . The computer-implemented method as recited in  claim 3 , further comprising transmitting the action report to at least one of the subject entity, a governmental agency, a financial institution, and a third-party entity. 
     
     
         5 . The computer-implemented method as recited in  claim 2 , further comprising determining that at least one of the subset of the attributes of the subject entity contributing most to the credit risk score is compliant with respect to a set of regulatory compliance rules. 
     
     
         6 . The computer-implemented method as recited in  claim 2 , wherein the next regression tree and the subsequent regression trees are generated using a gradient descent process. 
     
     
         7 . The computer-implemented method as recited in  claim 2 , further comprising constructing the next regression tree and the subsequent regression trees using only a subgroup of attribute variables. 
     
     
         8 . The computer-implemented method as recited in  claim 7 , wherein the subgroup of attribute variables is randomly selected from the set of attribute variables. 
     
     
         9 . The computer-implemented method as recited in  claim 2 , wherein the set of attribute variables comprise one or more socio-economic variables, demographic variables, medical history variables, financial history variables, and variables based on social media network profiles for the N entities. 
     
     
         10 . The computer-implemented method as recited in  claim 1 , further comprising determining credit risk scores for a plurality of subject entities and storing credit risk score component data for only respective subsets of the attribute variables corresponding to the subject entities to minimize storage resources. 
     
     
         11 . An apparatus, comprising:
 one or more processors; and   one or more persistent memory modules coupled to the one or more processors, the one or more persistent memory modules having program instructions stored thereon which, when executed by the one or more processors, are configured to perform following acts:
 configuring a Gradient Boosted Tree (GBT) ensemble comprising a plurality of regression trees as a trained decision engine based on a training dataset containing N records corresponding to N entities respectively, each record comprising a value relating to a performance variable and a plurality of values corresponding to a set of attribute variables; 
 receiving, over a communications network, input values corresponding to a plurality of attributes associated with a subject entity; 
 determining, by the trained decision engine, a credit risk score for the subject entity responsive to the input values; 
 decomposing the credit risk score into a set of linearly decomposed constituent components corresponding to the attributes of the subject entity; 
 identifying a subset of the attributes of the subject entity contributing most to the credit risk score; and 
 performing a decision-making action with respect to the subject entity based at least in part upon one or more respective portions of the credit risk score associated with the subset of attributes of the subject entity. 
   
     
     
         12 . The apparatus as recited in  claim 10 , wherein the program instructions are further configured to train the GBT ensemble, the training including:
 constructing an initial regression tree based on minimization of a cost function associated with a root node corresponding to a select attribute variable and iteratively branching a training population into sub-populations corresponding to a set of leaf nodes based on remaining attribute variables;   determining a residual value and a corresponding loss function associated with the initial regression tree;   constructing a next regression tree based on the residual value and the loss function of the initial regression tree; and   iteratively generating subsequent regression trees based on a predecessor regression tree's residual value and corresponding loss function associated therewith, the iterative generation continuing until a predetermined number of regression trees are obtained as a fitted ensemble of the plurality of regression trees operable as the trained decision engine.   
     
     
         13 . The apparatus as recited in  claim 12 , wherein the program instructions further include instructions for generating an action report regardless of an adverse decision based on the credit risk score. 
     
     
         14 . The apparatus as recited in  claim 13 , wherein the program instructions further include instructions for transmitting the action report to at least one of the subject entity, a governmental agency, a financial institution, and a third-party entity. 
     
     
         15 . The apparatus as recited in  claim 12 , wherein the program instructions further include instructions for determining that at least one of the subset of the attributes of the subject entity contributing most to the credit risk score is compliant with respect to a set of regulatory compliance rules. 
     
     
         16 . The apparatus as recited in  claim 12 , wherein the next regression tree and the subsequent regression trees are generated using a gradient descent process. 
     
     
         17 . The apparatus as recited in  claim 12 , wherein the program instructions further include instructions for constructing the next regression tree and the subsequent regression trees using only a subgroup of attribute variables. 
     
     
         18 . The apparatus as recited in  claim 17 , wherein the subgroup of attribute variables is randomly selected from the set of attribute variables. 
     
     
         19 . The apparatus as recited in  claim 12 , wherein the set of attribute variables comprise one or more socio-economic variables, demographic variables, medical history variables, financial history variables, and variables based on social media network profiles for the N entities. 
     
     
         20 . The apparatus as recited in  claim 12 , wherein the program instructions further include instructions for determining credit risk scores for a plurality of subject entities and storing credit risk score component data for only respective subsets of the attribute variables corresponding to the subject entities to minimize storage resources.