Consistency-based search in feature selection

摘要: Feature selection is an effective technique in dealing with dimensionality reduction. For classification, it used to find "optimal" subset of relevant features such that the overall accuracy classification increased while data size reduced and comprehensibility improved. methods contain two important aspects: evaluation a candidate feature search through space. Existing algorithms adopt various measures evaluate goodness subsets. This work focuses on inconsistency measure according which inconsistent if there exist at least instances same values but different class labels. We compare other study strategies as exhaustive, complete, heuristic random search, can be applied this measure. conduct empirical examine pros cons these methods, give some guidelines choosing method, classifier error rates before after selection.