A Novel Multi-Label Evaluation Measure with Comparative Analysis
EunSang Bak, Youngeun An, Sungbum Pan · 2023
Multi-label classification presents unique challenges that cannot be resolved by simply adapting single-label classification methods. One of the key challenges lies in devising an effective performance evaluation method. Although several evaluation measures for multi-label classification have been proposed, none of them has gained universal acceptance. This study focuses on ranking-based metrics and compares commonly used ranking-based methods in depth. Through this analysis, we identify the limitations of existing methods and leverage the characteristics of binary sequences to effectively overcome these drawbacks. The relationship between a ranking table and a binary sequence provides an efficient means to compare the characteristics of performance metrics, leading to the development of a novel measure for evaluating multi-label classification performance.