Evolutionary-Based Ensemble Under-Sampling for Imbalanced Data

• Published in: 2019 16th International Computer Conference on Wavelet Active Media Technology and Information Processing pp. 212 - 216
• Main Authors: ZHANG, YONGQING, LU, RONGZHAO, HUANG, JI, GAO, DONGRUI
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.12.2019
• Subjects: Classification algorithms; Ensemble learning; Evolutionary algorithms; Evolutionary computation; Forestry; Imbalanced data; Information technology; Machine learning; Machine learning algorithms; Sampling method; Sampling methods
• ISBN: 9781728142418; 1728142415
• ISSN: 2576-8964
• DOI: 10.1109/ICCWAMTIP47768.2019.9067647

Class imbalance is a prevalent problem in the real world, which mainly refers to the uneven distribution of samples of each class, and it will lead to a serious decline in the learning performance. The classification results will seriously bias to the majority class and ignore the minority class. However, the accuracy of the minority class is usually the focus of attention. Therefore, in this paper, an evolutionary-based ensemble under-sampling (EEU) algorithm is proposed to solve this problem. Specifically, evolutionary algorithm is used to under sample the data and multiple base classifiers are trained by ensemble learning. The advantage of this algorithm lies in that it can improve the accuracy of minority class. Comparison experiments are performed on five UCI datasets, and the results demonstrate that EEU outperforms other sampling methods.