×

Feature generation and model selection for generalized linear models

  • US 9,292,550 B2
  • Filed: 02/21/2013
  • Issued: 03/22/2016
  • Est. Priority Date: 02/21/2013
  • Status: Active Grant
First Claim
Patent Images

1. A computer-implemented method comprising:

  • identifying a dataset that stores values for a target attribute and input attributes, where the input attributes are under consideration for inclusion in a generalized linear model that predicts a value of the target attribute based on a selection of features, where each feature comprises a combination of one or more of the input attributes;

    identifying candidate features, where a candidate feature comprises a combination of one or more of the input attributes;

    computing respective inclusion scores for respective candidate features, based, at least in part on a likelihood that the candidate feature will be selected for inclusion in the generalized linear model;

    ordering the candidate features according to inclusion score;

    constructing a set of one or more branches of candidate features, where each branch includes candidate features ordered according to inclusion score from highest inclusion score to lowest inclusion score, where the one or more branches do not include candidate features having an inclusion score below a predetermined minimum score; and

    providing a branch of candidate features to a streamwise feature selection process configured to construct the generalized linear model by considering candidate features in the branch, in turn, starting with the candidate feature with the highest inclusion score, and including selected candidate features in the generalized linear model.

View all claims
  • 1 Assignment
Timeline View
Assignment View
    ×
    ×