Multi-label Classification with ART Neural Networks

Multi-label Classification (MC) is a classification task with instances labeled by multiple classes rather than just one. This task becomes increasingly important in such fields as gene function prediction or web-mining. Early approaches to MC were based on learning independent binary classifiers for each class and combining their outputs in order to obtain multi-label predictions. Alternatively, a classifier can be directly trained to predict a label set of an unknown size for each unseen instance. Recently, several direct multi-label learning algorithms have been proposed. This paper investigates a novel method to solve a MC task by using an Adaptive Resonance Theory (ART) neural network. A modified Fuzzy ARTMAP algorithm Multi-Label-FAM (ML-FAM) was applied to classification of multi-label data. The obtained preliminary results on the Yeast data set and their comparison with the results of existing algorithms demonstrate the effectiveness of the proposed approach.