An Efficient Approach for Multi-Label Streaming Feature Selection

With the rapid growth of multi-label streaming data, efficient feature selection becomes a critical challenge. Traditional methods often struggle to handle the dynamic nature of continuously arriving data. This paper introduces OSM-MI, a novel online feature selection method designed for multi-label streaming datasets. OSM-MI uses mutual information to dynamically select features, minimizing redundancy and maximizing relevance. The method is compared with existing algorithms, including OM-NRS, OMGFS, and MUCO, across several datasets such as Yeast, Medical, Scene, Enron, and others. Experimental results show that OSM-MI outperforms the other methods in terms of accuracy, precision, and efficiency, while also maintaining lower execution times. Statistical significance is confirmed through the Wilcoxon test, demonstrating OSM-MI's robustness for real-time multi-label classification. This work provides an efficient, scalable solution for feature selection in streaming environments.