Conditional Mutual Information Based Feature Selection

In this paper, a new greedy feature selection algorithm is proposed to detect more precisely informative features. It overcomes the limitation of many existing MI-based gready feature selection algorithms. It is capable of detecting the relation of relevant feature combinations in some degree.In addition, the requirements of the memory storage and computation cost are low. Experimental results for the UCI benchmark dataset demonstrate the good performance of the proposed algorithm on the experimented data sets.