Filtering Feature Selection Algorithm based on Fusion Strategy

This paper mainly concerns filtering algorithms for feature selection of high-dimensional data in the field of machine learning. Feature selection implies that some of data is useless or even plays a negative role for machine learning, more specifically, there are redundant and invalid features among them. This paper applies fusion strategy for feature selection by combining different filtering standards to obtain a new one. The basic architecture of step-by-step filtering is described in order to ensure the accuracy and efficiency of feature selection, upon which the Relief-F-MRMR filtering criterion and specific operation steps are designed. An illustrative example is provided to show the validity and advantage of the proposed approach.