Dynamic interactive weighted feature selection using fuzzy interaction information

Traditional information theory-based feature selection methods are designed for discrete features, which require additional discretization steps when working with continuous features. In contrast, fuzzy information theory-based feature selection methods can handle continuous features directly. However, most existing fuzzy information theory-based feature selection methods do not consider the dynamic interaction between candidate features and the already selected ones. To address this issue, we propose a dynamic weighted feature selection method based on fuzzy interaction information that can handle continuous features. First, we use fuzzy information theory metrics to characterize the concepts of feature relevance, redundancy, and interaction. Second, we define a fuzzy interaction weight factor that can quantify the redundancy and interaction between features by using fuzzy interaction information. Third, we design a novel feature selection algorithm called fuzzy dynamic interactive weighted feature selection (FDIWFS) by combining the fuzzy interaction weight factor with a sequential forward search strategy. To evaluate the effectiveness of FDIWFS, we compare it with eight state-of-the-art feature selection methods on fifteen publicly available datasets. The results of comparative experiments demonstrate that FDIWFS outperforms the other methods in terms of classification performance.