Ground truth bias in external cluster validity indices

详细信息    查看全文

• 作者：Yang Lei^a ; ^{yalei@student.unimelb.edu.au}Author Vitae ; James C. Bezdek^a ; ^{jbezdek@unimelb.edu.au}Author Vitae ; Simone Romano^a ; ^{simone.romano@unimelb.edu.au}Author Vitae ; Nguyen Xuan Vinh^a ; ^{vinh.nguyen@unimelb.edu.au}Author Vitae ; Jeffrey Chan^b ; ^{jeffrey.chan@rmit.edu.au}Author Vitae ; James Bailey^a ; ^{baileyj@unimelb.edu.au}Author Vitae
• 关键词：External cluster validity indices ; Rand index ; Ground truth bias ; Quadratic entropy
• 刊名：Pattern Recognition
• 出版年：2017
• 出版时间：May 2017
• 年：2017
• 卷：65
• 期：Complete
• 页码：58-70
• 全文大小：1464 K
• 卷排序：65

文摘

We identify the GT bias effect for external validation measures, and explain its importance. We test and discuss NC bias for 26 popular pair-counting based validation measures. We prove that the RI and related 4 indices suffer from GT bias. We provide theoretical explanations for understanding why and when GT bias happens. We present experimental results that support our analysis. We present an empirical example to show that the ARI also suffers from a modified GT bias.