Document Type : Original/Review Paper

Authors

• Ebrahim NezamParast Bonab ¹
• Vahid Ranjbar ²
• Mohammad ghasemzade ¹

¹ Department of Computer Engineering, Yazd University, Yazd, Iran

² Department of computer engineering, Yazd University, Yazd, Iran

Abstract

Unsupervised clustering of multi-dimensional data, particularly in the presence of complex topologies, remains a challenging problem in data analysis and machine learning. This is especially critical in the medical domain, where precise diagnosis and cluster separation are of paramount importance. Although meta-heuristic algorithms exhibit reasonable performance in this area, they frequently suffer from convergence stagnation, dimensional freezing, and the deterioration of cluster cohesion in high-dimensional search spaces. To overcome these limitations, this research proposes a novel hybrid algorithm integrating the Aquila Optimizer (AO) and Harris Hawks Optimization (HHO). Initially, a sliding-window mechanism is designed to accurately detect stagnation regions during the search process. Subsequently, a dual, rank-based surgical strategy is applied to the population. For average-quality particles, a targeted and constrained Cauchy mutation is applied exclusively to critical dimensions, whereas a structural crossover operator is employed for the weakest particles to reconstruct and recover the search space. Empirical evaluations conducted on five standard datasets—two of which are related to medical diseases—demonstrate that the proposed architecture significantly minimizes the sum of squared errors in complex spaces. These findings highlight the high efficacy of structural surgical strategies in enhancing exploration and exploitation for multi-dimensional optimization problems.

Keywords

• Unsupervised clustering
• Metaheuristic optimization
• Hybrid algorithms
• Stagnation recovery
• Dimensional freezing

Main Subjects

• H.6.4. Clustering