A Discernibility Matrix Based Algorithm for Feature Selection

An algorithm of reduct computation for feature selection is proposed in the paper, which is a discernibility matrix based method and aims at reducing the number of irrelevant and redundant features in data mining. The method used both significance information of attributes and information of discernibility matrix to define the necessity of heuristic feature selection. The advantage of the algorithm is that it can find an optimal reduct in most cases. Experimental results confirmed the above assertion and also shown that the proposed algorithm is more efficient in time performance