US 11,798,018 B2
Efficient feature selection for predictive models using semantic classification and generative filtering
Wei Zhang, Great Falls, VA (US); Shiladitya Bose, Kolkata (IN); Said Kobeissi, Lovettsville, VA (US); Scott Allen Tomko, Leesburg, VA (US); and Jeremy W King, Aldie, VA (US)
Assigned to Adobe Inc., San Jose, CA (US)
Filed by ADOBE INC., San Jose, CA (US)
Filed on Mar. 7, 2016, as Appl. No. 15/62,937.
Prior Publication US 2017/0255952 A1, Sep. 7, 2017
Int. Cl. G06N 20/00 (2019.01); G06Q 30/0204 (2023.01); G06Q 10/0639 (2023.01); G06Q 50/00 (2012.01); G06F 16/9535 (2019.01); G06F 16/35 (2019.01); G06N 3/08 (2023.01); G06N 3/044 (2023.01); G06N 3/045 (2023.01)
CPC G06Q 30/0205 (2013.01) [G06F 16/35 (2019.01); G06F 16/9535 (2019.01); G06Q 10/06395 (2013.01); G06Q 50/01 (2013.01); G06N 3/044 (2023.01); G06N 3/045 (2023.01); G06N 3/08 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A computer-implemented method for generating a predictive model using a reduced feature set, the method comprising:
obtaining, by a one or more processors, a set of features;
classifying, by the one or more processors executing a semantic classifier, features of the set of features into feature subsets corresponding with semantic classes, the semantic classes being provided in a ranked order;
selecting, by the one or more processors, features from the feature subsets to include in the reduced feature set, the features for the reduced feature set being selected by iterating through at least a portion of the feature subsets in the ranked order of the semantic classes, each iteration of a feature subset performed by the one or more processors:
computing, by a generative filter of the one or more processors, a feature quality score for each feature in the feature subset to identify low quality features,
removing, by the generative filter of the one or more processors, the low quality features from the feature subset to provide a filtered feature subset, and
performing, by a forward selection module of the one or more processors, forward selection on the filtered feature subset, wherein the forward selection considers selected features from a previously processed filtered feature subset;
outputting, by the one or more processors, the reduced feature set that includes the features selected using the forward selection; and
generating, by the one or more processors, the predictive model using the features included in the reduced feature set.