Feature Selection Based on Cloud Theory

Feature selection (FS) can effectively improve the speed and accuracy of classification. The traditional FS approaches usually score a single feature, do not evaluate feature subset. Through constructing cloud model for every feature in every class, a method for measuring feature´s classification ability was proposed, and feature supplement degree was defined to describing the relationship between a single feature and feature subset. And then, two feature selection algorithms, Tactic 1 and Tactic2 were designed based on feature´s classification ability and feature supplement capability. Experiments in six data sets of UCI show: compared with other traditional feature selection approaches: SFFS, SA and NN, the classification precision obtained by Tacticl is almost equivalent to AS, Tactic2 is almost equivalent to NN, but Tactic 1 and Tactic2 can get better generalization performance. Especially, Tactic2 is no need to initialize the number of feature; Furthermore, Tactic 1 and Tactic2 have much superiority in computation complexity when the number of classes of dataset is very small comparing with the number of features.