A Fast Feature Selection Method Based on Mutual Information in Multi-label Learning

• Published in: Computer Supported Cooperative Work and Social Computing Vol. 917; pp. 424 - 437
• Main Authors: Sun, Zhenqiang, Zhang, Jia, Luo, Zhiming, Cao, Donglin, Li, Shaozi
• Format: Book Chapter
• Language: English
• Published: Singapore Springer 2019; Springer Singapore
• Series: Communications in Computer and Information Science
• Subjects: Feature selection; Label-correlation; Multi-label learning; Mutual information
• ISBN: 9789811330438; 9811330433
• ISSN: 1865-0929; 1865-0937
• DOI: 10.1007/978-981-13-3044-5_31

Recently, multi-label learning is concerned and studied in lots of fields by many researchers. However, multi-label datasets often have noisy, irrelevant and redundant features with high dimensionality. Accompanying with these issues, a critical challenge called “the curse of dimensionality” makes many tasks of multi-label learning very difficult. Therefore, many method such as feature selection to solve this problem has received much attention. Among many feature selection methods, a large number of information-theoretical-based methods are developed to solve the learning issue and the results are very good. Unfortunately, most of existing feature selection methods are either directly transformed from single-label methods or insufficient in light of using heuristic algorithms as the search component. Motivated by this, a novel fast method based on mutual information with no parameter is proposed, which obtains the optimal solution via constrained convex optimization with less time. Specifically, by incorporating the label information into the feature selection process, label-correlation is taken into consideration to generate the generalized model.