A feature selection method driven by fuzzy implication granularity

Abstract

The data in real life are often very complex, with different types and scales, and contain a large number of redundant features. How to perform feature selection for complex data is a tricky problem. To address the issue, this paper proposes a filter-based feature selection method driven by fuzzy implication granularity (FIGFS). Firstly, the fuzzy adaptive neighborhood radius is proposed to construct the information granules, and on this basis, a series of multi-granularity fuzzy implication information measures are established to characterize the feature uncertainty. Secondly, granular consistency is proposed to capture the correlation between features and decisions at the overall and local levels respectively. Then, a new multi-criteria feature evaluation metric is constructed by combining granularity consistency and multi-granularity fuzzy implication information measures. Finally, a general forward search feature selection algorithm compatible with low-dimensional data and high-dimensional data is designed. Compared with six state-of-the-art algorithms on 24 public datasets, the results show that our method is feasible and superior.