A mutual information based feature selection algorithm

The objective of the eliminating process is to reduce the size of the input feature set and at the same time to retain the class discriminatory information. This paper proposes and evaluates a new feature selection algorithm using information theory which is the mutual information (MI) between combinations of input features and the class instead of mutual information between a single input feature and the class for both continuous-valued and discrete-valued features. Comparison studies of new and previously published classification algorithms indicate that the proposed algorithm is robust, stable and efficient.