Multi-label Classification for Handling Mixed Data via Adaptive Resonance Theory-based Clustering

Abstract

Various multi-label classifiers have been proposed for multi-label classification problems. Our previous study has proposed an Adaptive Resonance Theory (ART)-based clustering method for the multi-label classification problems, called Multi-Label CIM-based ART (MLCA). MLCA adaptively and continually generates nodes corresponding to input data, and the generated nodes are used as a classifier. Many real-world multi-label classification problems are mixed datasets that contain both numerical and categorical attributes. In a mixed dataset, it is necessary to apply an appropriate similarity for each numerical and categorical attribute. This study extends MLCA to mixed datasets by measuring the similarity between data using correntropy for numerical attributes and hamming distance for categorical attributes. Numerical experiments on real-world datasets show the effectiveness of the proposed method.