Research on Unbiased Probability Estimation by Equidistant ECOC

摘要：

By error-correcting output codes(ECOC) is a useful way to solve multi-class classification problem, and how to get the probability output of a multi-class classifier has become obvious important for many applications. How to get the accurate probability estimation via equidistant ECOC is investigated in this paper. The constraint conditions are reasoned out for equidistant ECOC to get unbiased probability estimation, and the classification performance of different equidistant ECOC are investigated and the experimental results and analysis are given.

关键词： Multi-class classification Error-correcting output codes Probability estimation

作者: Xiaodan Wang Jindeng Zhou

作者单位: Department of Computer Engineering, Air Force Engineering University

会议类型: 国际会议

会议名称: 2010 International Conference on Information Security and Artificial Intelligence(2010年信息安全与人工智能国际会议 ISAI 2010)

会议地点: 成都

会议语种:英文

页码: 301-305

在线出版日期: 2010-12-17（万方平台首次上网日期，不代表论文的发表时间）