A Two-Phase Heuristic Construction of Feature Sets for Classification

Abstract

The aim of feature selection applied to a classification task is to find a minimal subset of features for being used in the classification. Some researches have focused their effort on selecting a useful set of attributes, others on selecting a relevant and not redundant set of attributes. We proposed a heuristic construction algorithm for selecting a useful and not redundant subset of features. The algorithm proposed belongs to the filter approach and make use of a correlation measure for the task.