Gradient-based elephant herding optimization for cluster analysis.

Area of Science:

• Data Science and Machine Learning
• Computational Intelligence
• Optimization Algorithms

Background:

• Traditional clustering methods often get stuck in local optima and are sensitive to initial conditions.
• Existing algorithms struggle with balancing exploration and exploitation phases in complex datasets.

Purpose of the Study:

• To introduce a novel clustering algorithm, gradient-based elephant herding optimization (GBEHO), to address limitations of traditional methods.
• To enhance the initial centroid selection and improve the exploration-exploitation balance in clustering.
• To evaluate the performance and stability of GBEHO against other metaheuristic algorithms.

Main Methods:

• Developed GBEHO by integrating the elephant optimization algorithm (EHO) with the gradient-based optimizer (GBO) for initial centroid selection.
• Incorporated Gaussian chaos mapping to improve population initialization and balance exploration/exploitation.
• Introduced random wandering and variation operators to refine agent location update strategies.
• Evaluated GBEHO on nine synthetic and real-world datasets using accuracy rate, specificity, detection rate, and F-measure.

Main Results:

• GBEHO achieved the top rank among 10 evaluated algorithms.
• The proposed method demonstrated superior performance compared to state-of-the-art techniques.
• GBEHO exhibited excellent accuracy, specificity, detection rate, and F-measure.
• The algorithm showed enhanced stability in clustering tasks.

Conclusions:

• GBEHO offers a robust and effective solution for cluster analysis, overcoming local optima and initial solution dependency.
• The novel integration of gradient-based optimization and enhanced EHO provides significant improvements in clustering performance.
• GBEHO presents a promising advancement in metaheuristic-based clustering with superior accuracy and stability.