Application of an Improved Multimodal Multiobjective Algorithm in Feature Selection.

Feature selection is the process of screening the original data set and selecting the most effective features to reduce the data dimension. In the process of feature selection, when selecting the same number of features, there may be different combinations of features, making the final obtained classification results very similar. This is a multimodal multiobjective (MMO) feature selection optimization problem. In this paper, an improved multimodal multiobjective evolutionary algorithm based on uniform crossover is proposed for MMO feature selection optimization problem. The algorithm uses the strategy based on the niche of decision space and the uniform crossover strategy, which has better global optimization ability and can find more equivalent feature subsets. The algorithm is tested on seven feature selection datasets and compared with the existing multimodal multiobjective feature selection algorithms. Experimental results show that the algorithm can find more equivalent feature combinations without reducing the classification accuracy.