An Enhanced Feature Selection Algorithm Based on Maximum Relevance Minimum Redundancy and Splicing Strategy¹

To address the challenge of complexity and the difficulty in determining the number of optimal feature subsets in feature selection, this paper proposed an enhanced feature selection algorithm which can be used when preprocessing data in machine learning projects to improve the performance and efficiency of models. This method integrates the principles of maximum relevance and minimum redundancy with the splicing technique. This study also integrates the relationship between the number of feature subsets and the error rate, introducing weighted values to feature subset count and error rates. Experimental result conducted on UCI datasets demonstrates the method's capability to effectively eliminate irrelevant redundant features while maintaining accuracy. The approach significantly reduces the number of features.