Affiliations: [a] School of Management, Hefei University of Technology, Hefei, China
| [b] Key Laboratory of Process Optimization and Intelligent Decision-making, Ministry of Education, Hefei, China
| [c] Business School, Huzhou University, Huzhou, China
| [d] School of Management Science and Engineering, Anhui University of Finance and Economics, Bengbu, China
Correspondence:
[*]
Corresponding author. Zhiwei Ni. E-mail: zhiwein@163.com.
Abstract: Diversity and accuracy of classifiers are widely recognized to be two key factors for a successful ensemble. The increase of diversity among classifiers must lead to the decrease of the average accuracy of that, and vice verse. Therefore, finding a tradeoff between the diversity and the accuracy of classifiers can make the ensemble perform the best. Existing ensemble pruning approaches always find the tradeoff using diversity measures and heuristic algorithms separately. Those ensemble pruning approaches based on diversity measures, using different strategies, cannot exactly find the tradeoff; Those approaches based on heuristic algorithms cannot also exhaustively search for that. To address the issue, Combining Weak-link Co-evolution Binary Artificial Fish swarm algorithm and Complementarity measure for Ensemble Pruning (CWCBAFCEP) is proposed using a combination of the proposed Weak-link Co-evolution Binary Artificial Fish Swarm Algorithm (WCBAFSA) and COMplementarity measure (COM). First, the classifiers in a constructed initial pool of classifiers are pre-pruned using COM, which significantly reduce the computational complexity of ensemble pruning. Second, the final ensemble extracted from the remaining classifiers after pre-pruning can be efficiently achieved using the proposed WCBAFSA. Experimental results on 25 datasets from the UCI Machine Learning Repository demonstrate that CWCBAFCEP performs much better than the original ensemble and other state-of-the-art ensemble pruning approaches, and that its effectiveness and efficiency. It provides a new research idea for ensemble pruning.
