A new ensemble method for multi-label data stream classification in non-stationary environment

Abstract

Most existing approaches for the data stream classification focus on single-label data in non-stationary environment. In these methods, each instance can only be tagged with one label. However, in many realistic applications, each instance should be tagged with more than one label. To address the challenge of classifying multi-label stream in evolving environment, we propose a novel Multi-Label Dynamic Ensemble (MLDE) approach. The proposed MLDE integrates a number of Multi-Label Cluster-based Classifiers (MLCCs). MLDE includes an adaptive ensemble method and an ensemble voting method with two important weights, subset accuracy weight and similarity weight. Experimental results reveal that MLDE achieves better performance than state-of-the-art multi-label stream classification algorithms.