AN ENSEMBLE ANOMALY DETECTION WITH IMBALANCED DATA BASED ON ROBOT VISION

作者：Wang, Yongxiong^*; Sun, Shuxin^*; Zhong, Jiandong^*

来源：International Journal of Robotics and Automation, 2016, 31(2): 77-83.

DOI：10.2316/Journal.206.2016.2.206-4095

• Anomaly detection
• robot vision
• ensemble learning
• imbalanced data
• Tabu search

One of the main causes leading to low recognition rates in anomaly detection is to neglect the imbalance of the data distribution. An ensemble algorithm based on robot vision is proposed, which combines multiple semi-supervised k-means methods and C4.5 to relieve the problems of imbalanced data in anomaly detection. Tabu search with cost-sensitive function is simultaneously applied to select features and weights of detectors. The multiple k-means sub-detectors first classify the feature space with the diversity of features or multi-subclasses into k clusters. Then C4.5s improve the detection accuracy by weighted summing the results of the two machine learning methods. Experimental results on real-world data with severe imbalance and multi-subclasses demonstrate that the proposed method is effective in classifying imbalanced data.

• 出版日期2016
• 单位上海理工大学; 上海交通大学