Exclusive Lasso Assisted Two-Stage Fuzzy-Rough Feature Selection

The application of fuzzy-rough sets (FRS) in feature selection (FS) has employed the dependency degree to guide the FS process with much success. Whilst promising, most existing fuzzy-rough feature selection (FRFS) approaches are only conducted at the level of individual features, considering the inclusion/exclusion of individual features with regard to a candidate feature subset. This can overlook important information about inherent feature structures, such as correlations between features or their collaborative contributions to a common decision. To address this issue, an exclusive lasso assisted two-stage fuzzy-rough FS (EL-TSFRFS) method is presented in this paper. The approach first partitions features into groups using K-means clustering, then applies exclusive lasso regularisation to select and rank representative features within each group. In the second stage, a fuzzy-rough feature selection algorithm is employed at the group level to determine the final discriminative feature subset. Comprehensive experiments on a diverse range of datasets demonstrate that EL-TSFRFS consistently produces smaller feature subsets while achieving superior or comparable classification accuracy compared to state-of-the-art FRFS techniques. The proposed method also exhibits robust performance across different classifiers and parameter settings, making it a promising tool for dimensionality reduction and knowledge discovery in complex data analysis tasks.