Evolutionary Clustering

This research area focuses on developing evolutionary algorithm (EA) methods that can automatically determine or adapt the number of clusters k during the clustering process, addressing limitations of classical K-means and other fixed-k algorithms. It tackles challenges like unknown cluster count, stream data dynamics, and the NP-hardness of clustering, employing multi-objective and metaheuristic strategies to balance cluster compactness and separation. This theme is vital because real-world data often lack prior knowledge of cluster numbers, and adaptive clustering enhances clustering robustness, quality, and practical applicability.

This theme addresses the crucial challenge of choosing a single best clustering solution from the set of nondominated trade-off solutions yielded by multi-objective evolutionary clustering (EMC) algorithms. Since EMC produces multiple Pareto-optimal partitions balancing conflicting criteria (e.g., cohesion vs separation), proper decision-making methods must evaluate and rank these solutions based on problem-specific context and quality. Research in this area explores machine learning-based decision-making, consensus methods, and geometric approaches to improve robustness and generalization, enabling practical application of EMC.

This research area investigates combining evolutionary algorithms (EAs), such as genetic algorithms (GAs), swarm intelligence, and black hole algorithms, with classical clustering methods like K-means to boost clustering optimization. Hybrid approaches address common issues of classical techniques (e.g., local optima, initial centroid sensitivity, fixed k) by leveraging heuristic global search and bio-inspired operators to improve convergence, solution quality, and applicability to complex or large-scale data, including image segmentation and high-dimensional data. This theme bridges optimization theory with practical applications in unsupervised learning.