Classification is a classic technique employed in data mining. Many ensemble learning methods have been introduced to improve the predictive accuracy of classification. A typical ensemble learning method consists of three steps: selection, building, and integration. Of the three steps, the first and third significantly affect the predictive accuracy of the classification. In this paper, we propose a new selection and integration scheme. Our method can improve the accuracy of subtrees and maintain their diversity. Through a new voting scheme, the predictive accuracy of ensemble learning is improved. We also theoretically analyzed the selection and integration steps of our method. The results of experimental analyses show that our method can achieve better accuracy than two state-of-the-art tree-based ensemble learning approaches.
All Science Journal Classification (ASJC) codes
- Theoretical Computer Science
- Geometry and Topology