Drift Detection for Multi-label Data Streams Based on Label Grouping and Entropy

Abstract

Many real-world applications involve multi-label data streams, so effective concept drift detection methods should be able to consider the unique properties of multi-label stream data, such as label dependence. To deal with these challenges, we proposed an efficient and effective method to detect concept drift based on label grouping and entropy for multi-label data. Two methods are proposed to group the set of class labels into different subsets and a multi-label version of entropy was adjusted to measure the distribution of multi-label data. Concept drift was detected by comparing the entropies of the older and the most recent data. The experiments are run on three synthetic datasets and two real-world datasets and the experimental results illustrate the better classification performance of the proposed method for detecting drift.