Dimension reduction in predictive model development

Abstract

Models are generated using a variety of tools and features of a model generation platform. For example, in connection with a project in which a user generates a predictive model based on historical data about a system being modeled, the user is provided through a graphical user interface a structured sequence of model generation activities to be followed, the sequence including dimension reduction, model generation, model process validation, and model re-generation. Historical multi-dimensional data is received representing multiple variables to be used as an input to a predictive model of a commercial system variables are pruned for which the data is sparse or missing, and the population of variables is adjusted to represent main effects exhibited by the data and interaction and non-linear effects exhibited by the data.

1 . A medium storing instructions that are executable by a processor to:
 receive historical multi-dimensional data representing multiple variables; and   adjust a population of variables to represent interaction effects exhibited by the historical multi-dimensional data,
 the population of variables being selected from the multiple variables for use in generating a predictive model, at least some of the multiple variables being excluded from the selected population of variables, 
 the interaction effects represented by the adjusted population of variables including interactions of at least some of the excluded variables with each other and with variables in the selected population of variables. 
   
     
     
         2 . The medium of  claim 1  in which the interaction effects include stages of main effect interactions. 
     
     
         3 . The medium of  claim 1  in which the instructions are executable by the processor to:
 transform one or more of the multiple variables into one or more predictive variables, the transformation comprising adjusting a response frequency associated with each of the one or more of the multiple variables by a Bayesian analysis based on a priori response frequency associated with the variable. 
 
     
     
         4 . The medium of  claim 3  in which adjusting the response frequency comprises associating the variable with a weight to regress the response frequency toward a mean response frequency. 
     
     
         5 . The medium of  claim 1  in which the instructions are executable by the processor further to:
 use the adjusted population of variables in generating the predictive model for use in interacting with a commercial system. 
 
     
     
         6 . The medium of  claim 1  in which the predictive model predicts behavior of a current or a prospective customer with respect to retention of a current service or product of a vendor. 
     
     
         7 . The medium of  claim 1  in which the predictive model predicts behavior of a current or a prospective customer with respect to risk of asserting claims, loan payment or prepayment to a vendor. 
     
     
         8 . The 1 medium of  claim 1  in which the predictive model predicts behavior of a current or a prospective customer with respect to usage of a current service or product of a vendor. 
     
     
         9 . The medium of  claim 1  in which the predictive model predicts behavior of a prospective or a current customer with respect to purchase of a product or service of a vendor. 
     
     
         10 . The medium of  claim 1  in which the instructions are executable by the processor to:
 generate the predictive model, and 
 enable the user to control staging of a sequence of activities in generating the predictive model through a user interface. 
 
     
     
         11 . The medium of  claim 1  in which the instructions are executable by the processor to:
 generate the predictive model, and 
 enable a user to reconstruct a sequence of choices involved in generating the predictive model. 
 
     
     
         12 . The medium of  claim 1  in which the instructions are executable by the processor to:
 enable a user to interactively manage two or more steps for adjusting the population of variables through a graphical user interface, the graphical user interface including an activation portion, which upon activation, enables the user to revisit at least one of the two or more steps. 
 
     
     
         13 . The medium of  claim 1  in which the historical multi-dimensional data comprises transaction and communication data, demographic data, and econometric data. 
     
     
         14 . The medium of  claim 1  in which the instructions are executable by the processor to:
 generate the predictive model, and 
 provide access to a user to one or more steps, including adjusting the population of variables used in generating the predictive model using a web service or Internet. 
 
     
     
         15 . A computer system comprising:
 a processor;   a memory; and   a storage device that stores a program for execution by the processor using the memory, the program comprising instructions for causing the processor to   receive historical multi-dimensional data representing multiple variables; and   adjust a population of variables to represent interaction effects exhibited by the historical multi-dimensional data,
 the population of variables being selected from the multiple variables for use in generating a predictive model, at least some of the multiple variables being excluded from the selected population of variables, 
 the interaction effects represented by the adjusted population of variables including interactions of at least some of the excluded variables with each other and with variables in the selected population of variables. 
   
     
     
         16 . The computer system of  claim 15  in which the program comprises instructions for causing the processor to:
 transform one or more of the multiple variables into one or more predictive variables, the transformation comprising associating each of the one or more of the multiple variables with a weight to regress a response frequency associated with the variable towards a mean response frequency. 
 
     
     
         17 . The computer system of  claim 15  in which the program comprises instructions for causing the processor to be accessible from a web service or Internet. 
     
     
         18 . A machine-based method comprising:
 receiving historical multi-dimensional data representing multiple variables; and   adjusting a population of variables to represent interaction effects exhibited by the historical multi-dimensional data,
 the population of variables being selected from the multiple variables for use in generating a predictive model, at least some of the multiple variables being excluded from the selected population of variables, 
 the interaction effects represented by the adjusted population of variables including interactions of at least some of the excluded variables with each other and with variables in the selected population of variables. 
   
     
     
         19 . The machine-based method of  claim 18  further comprising:
 transforming one or more of the multiple variables into one or more predictive variables, the transforming comprising associating each of one or more of the multiple variables with a weight to regress a response frequency associated with the variable towards a mean response frequency. 
 
     
     
         20 . The machine-based method of  claim 18  further comprising:
 generating the predictive model, and 
 enabling a user to access one or more steps, including adjusting a population of variables, used in generating the predictive model using a web service or Internet.