Comparative Analysis of State-of-the-Art Multi-Label Feature Selection Approaches for High-Dimensional Data

In today's world, technologies from various fields such as healthcare, social media, e-commerce, finance & banking, agriculture, education, etc., generate different types of data in massive quantities. Each sample of these data may have a large number of features and may have different types of class labels. Due to different types of class labels, these high-dimensional data can be categorized into single-label and multi-label data. From last few years, feature selection for multi-label classification has become a remarkable field of research. In this paper, we have reviewed several latest multi-label feature selection approaches focuses on the empirical analysis of multi-label classification task. A thorough experimental evaluation on five state-of-the-art multi-label feature selection methods conducted on seven numbers of benchmark multi-label datasets of multiple domains. Experimental results show that Correlation Label Enhancement-based Feature Selection (CLE-FS), as the most reliable choice for multi-label feature selection across various application domains.