Feature selection based on the combination of genetic algorithm and mutual information formula

In order to effectively reduce the feature dimension and obtain the optimal subset of features, a method for feature selection combining the genetic algorithm with the improved mutual information formula is proposed. The improved mutual information formula is used as the fitness function of genetic algorithm for feature selection. The algorithm presented in this paper combines the advantages of filter and wrapper method. The other two feature selection algorithms are used to compare with this method; and probabilistic neural network and BP neural network are used as classifiers to test these three types of feature selection algorithms. The effects of different approaches are compared by classification accuracies. Experimental results show that the proposed feature selection method can select features effectively with better generalized property.