Adaptive Resonance Theory-based Clustering with a Topological Structure for Multi-label classification

Abstract

Various multi-label classifiers have been proposed for multi-label classification problems. Among conventional studies, Multi-Label CIM-based ART (MLCA) has shown superior classification performance by realizing continual learning. MLCA adaptively and continually generates prototype nodes corresponding to given data, and the generated nodes are used as classifiers. Moreover, MLCA learns label information for each class independently and continually. Since label learning is performed by referring to the label information of neighbor nodes, the choice of neighbor nodes has a significant impact on the classification performance. However, there is a possibility to refer to excessive and harmful label information because MLCA always refers to a predefined number of neighbor nodes during learning. This paper introduces a method to MLCA that generates edges between nodes. The proposed algorithm defines neighboring nodes based on connected nodes by edges, which realizes an efficient and effective approach for learning label information. Experiment results show that the proposed algorithm has superior classification performance compared with conventional algorithms.