부산대학교 도서관

The feature selection problem involves selecting a subset of relevant features to enhance the performance of machine learning models, crucial for achieving model accuracy. Its complexity arises from the vast search space, necessitating the application of metaheuristic methods to efficiently identify optimal feature subsets. In this work, we employed a recently proposed metaheuristic algorithm named the Great Wall Construction Algorithm to address this challenge – a powerful optimizer with promising results. To enhance the algorithm’s performance in terms of exploration, exploitation, and avoidance of local optima, we integrated opposition-based learning and Gaussian mutation techniques. The proposed algorithm underwent a comprehensive comparative analysis against ten influential state-of-the-art methodologies, encompassing seven contemporary algorithms and three classical counterparts. The evaluation covered 22 datasets of varying sizes, ranging from 9 to 856 features, and included the utilization of six distinct evaluation metrics related to accuracy, classification error rate, number of selected features, and completion time to facilitate comprehensive comparisons. The obtained numerical results underwent rigorous scrutiny through several non-parametric statistical tests, including the Friedman test, the post hoc Dunn’s test, and the Wilcoxon signed ranks test. The resulting mean ranks and p-values unequivocally demonstrate the superior efficacy of the proposed algorithm in addressing the feature selection problem. The Matlab source code for the proposed approach is available for access via the link “https://www.mathworks.com/matlabcentral/fileexchange/159728-an-opposition-based-gwca-for-thefs-problem”.