Clustering Performance Analysis Using Chaotic and Lévy Flight-Enhanced Black-Winged Kite Algorithms.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Clustering is a key unsupervised learning method for pattern discovery in unlabeled data.
• Metaheuristic algorithms offer clustering solutions but often face premature convergence and low population diversity.
• Centroid-based clustering, often framed as Sum of Squared Errors (SSE) minimization, benefits from improved optimization techniques.

Purpose of the Study:

• To enhance metaheuristic clustering algorithms by addressing limitations like premature convergence and lack of diversity.
• To introduce and evaluate the effectiveness of the Chaotic Levy BKA (CLBKA) for centroid-based clustering.
• To investigate the impact of chaotic dynamics and Levy flight on clustering accuracy and stability.

Main Methods:

• The study utilizes the Black-Winged Kite Algorithm (BKA) and its enhanced versions: Chaotic BKA (CBKA), Lévy Flight-based BKA (LBKA), and Chaotic Levy BKA (CLBKA).
• Chaotic logistic mapping and Levy flight mechanisms are integrated to improve search diversity, adaptability, and long-range exploration.
• Cauchy-based perturbations are incorporated to enhance convergence stability.
• Algorithms are tested on sixteen UCI benchmark datasets with 30 independent runs, varying population and iteration settings.

Main Results:

• The Chaotic Levy BKA (CLBKA) demonstrated consistently superior clustering performance, achieving higher accuracy and stability.
• CLBKA achieved the lowest mean rank across various configurations in statistical validation using Friedman and Wilcoxon tests.
• The integration of chaotic dynamics and Levy flight significantly enhanced clustering robustness and optimization efficiency compared to other variants.

Conclusions:

• The developed Chaotic Levy BKA (CLBKA) effectively overcomes the limitations of traditional metaheuristic clustering algorithms.
• Integrating chaotic dynamics and Levy flight mechanisms is a promising strategy for improving clustering robustness and optimization.
• CLBKA offers a more accurate and stable approach for uncovering hidden patterns in unlabeled data.