A new feature selection approach based on new multi-exhaustive search

Feature Selection (FS) improves the classification model performance and comprehensibility by selecting informative feature subsets. It involves feature evaluators and search methods. Exhaustive search (ES) guarantees finding the optimal subset, but is computationally unfeasible, especially with large datasets. To overcome this drawback and take advantage of its optimality, this paper proposes a multi-exhaustive search (MES) to find the best subset in very low computational time. As a result, a new FS approach based on MES (FS-MES) is also proposed. The performance of FS-MES was evaluated using 11 benchmark datasets and compared with 13 FS approaches implemented in WEKA. A comparison with ES-based FS (FS-ES) demonstrated that FS-MES significantly reduced computation time and evaluations while maintaining competitive accuracy. Compared to the 12 other FS methods, FS-MES effectively outperformed all other FS methods in most of the datasets in terms of classification accuracy and number of selected features.