A Novel Swarm Intelligent Based Feature Selection Method for Stream Clustering

Abstract

Stream clustering faces challenges due to data dynamics, high dimensionality, and absence of labels. Existing feature selection methods rely on labeled data and threshold values, which limit their effectiveness on large, unstructured streaming data. To overcome these issues, we propose a swarm intelligence-based feature selection method for streaming environments, inspired by the problem-solving mechanism of a swarm. The Grey Wolf Algorithm is swarm intelligent based method known for its simplicity, minimal parameter derivation, and hunting behavior, which balances exploration and exploitation in complex search spaces. Our enhancements—including dynamic scaling, separate search phases, and retention of elite solutions—address premature convergence, prevent trapping in local optima, and reduce parameter sensitivity in the standard Grey Wolf Optimization algorithm. Our approach effectively selects semantically relevant features, achieving high clustering quality as evidenced by the Dunn Index (39.45–42.26), CH Index (0.678–0.893), and Davies-Bouldin Index (0.678–0.927) on different standard dataset used for clustering. The method proves to be robust, efficient, and suitable for real-time, unsupervised streaming data analysis, representing a significant advancement in dynamic feature selection.