A Filter Approach Based on Binary Integer Programming for Feature Selection
Bui Thi Mai Anh, Duong Viet Anh, Bui Quoc Trung · 2022 RIVF International Conference on Computing and Communication Technologies (RIVF) · 2022
Feature Selection (FS) aims to identify the most relevant information from the data in order to boost the overall performance of the learning model in terms of both accuracy and computational cost. Traditional FS methods rely on specific criteria to rank the features in order of their relevant scores and filter out low-ranking elements. However, the nature of these greedy approaches may suffer from the locally optimal choice. To tackle this problem, we propose a novel FS approach by reformulating the existing filter selection problem into an integer program which can be solved effectively by branch-and-cut algorithms to find the globally optimal solution (an optimal subset of features). Experiments on a wide range of benchmark datasets indicate that our proposed model outperforms existing state-of-the-art approaches in terms of considerably reducing the search space and computational resources.