<p>Unsupervised machine learning often faces challenges in achieving high-quality clustering, which involves grouping of data points into a predetermined number of clusters. In this paper, we have proposed a novel hybrid approach that combines the African Vultures Optimization Algorithm (AVOA) with the Harris hawks Optimizer (HHO) for data clustering. We introduce the new hybrid method, referred to as AVHH (AVOA + HHO), designed for efficient data clustering. In any optimization technique, exploration and exploitation are the two primary components. Exploration aims to discover promising cluster solutions within the given search space, while exploitation focuses on refining the clusters that have already been identified. In this proposed method, AVOA is used for exploration, while HHO is employed for exploitation. A detailed performance comparison was carried out for the proposed AVHH method against seven other state-of-the-art optimization algorithms, including AVOA, HHO, butterfly optimization algorithm (BOA), black hole algorithm (BHA), gray wolf optimizer (GWO), salp swarm algorithm (SSA), and sine cosine algorithm (SCA). The tests were conducted with 18 standard benchmark datasets often used in clustering, and key performance indicators are illustrated through the box plots, the convergence curves, and the sum of squared Euclidean distances. To evaluate the effectiveness of the proposed AVHH approach, rigorous statistical tests were performed, and the experimental results show that the proposed AVHH hybrid framework enhances the theoretical development of adaptive metaheuristics by linking exploration and exploitation controls with convergence stability, ultimately achieving superior clustering efficiency and accuracy across diverse datasets. To more rigorously evaluate the performance of the proposed AVHH algorithm on highly imbalanced and high-dimensional data, three benchmark datasets (SECOM, COIL20, and Lung Cancer) were utilized. The clustering performance of the proposed method is also evaluated on seven benchmark datasets using normalized mutual information (NMI), adjusted rand index (ARI), and accuracy (Acc) and is compared with other state-of-the-art algorithms. In addition, the computational run time (RT) of all considered methods are separately analyzed to assess their efficiency. To enhance transparency and reproducibility, the implementation of the proposed AVHH algorithm, along with the experimental configuration details, is available at <a href="https://github.com/tribhuvansingh88/AVHH-MATLAB-Code">https://github.com/tribhuvansingh88/AVHH-MATLAB-Code</a>.</p>

Hybridizing AVOA and HHO for robust data clustering: a novel optimization framework

期刊论文
发表日期：

2026年5月18日

查看全文

R. Janarthanan,
Tribhuvan Singh,
Subrat Kumar Nayak,
Debahuti Mishra

摘要

Unsupervised machine learning often faces challenges in achieving high-quality clustering, which involves grouping of data points into a predetermined number of clusters. In this paper, we have proposed a novel hybrid approach that combines the African Vultures Optimization Algorithm (AVOA) with the Harris hawks Optimizer (HHO) for data clustering. We introduce the new hybrid method, referred to as AVHH (AVOA + HHO), designed for efficient data clustering. In any optimization technique, exploration and exploitation are the two primary components. Exploration aims to discover promising cluster solutions within the given search space, while exploitation focuses on refining the clusters that have already been identified. In this proposed method, AVOA is used for exploration, while HHO is employed for exploitation. A detailed performance comparison was carried out for the proposed AVHH method against seven other state-of-the-art optimization algorithms, including AVOA, HHO, butterfly optimization algorithm (BOA), black hole algorithm (BHA), gray wolf optimizer (GWO), salp swarm algorithm (SSA), and sine cosine algorithm (SCA). The tests were conducted with 18 standard benchmark datasets often used in clustering, and key performance indicators are illustrated through the box plots, the convergence curves, and the sum of squared Euclidean distances. To evaluate the effectiveness of the proposed AVHH approach, rigorous statistical tests were performed, and the experimental results show that the proposed AVHH hybrid framework enhances the theoretical development of adaptive metaheuristics by linking exploration and exploitation controls with convergence stability, ultimately achieving superior clustering efficiency and accuracy across diverse datasets. To more rigorously evaluate the performance of the proposed AVHH algorithm on highly imbalanced and high-dimensional data, three benchmark datasets (SECOM, COIL20, and Lung Cancer) were utilized. The clustering performance of the proposed method is also evaluated on seven benchmark datasets using normalized mutual information (NMI), adjusted rand index (ARI), and accuracy (Acc) and is compared with other state-of-the-art algorithms. In addition, the computational run time (RT) of all considered methods are separately analyzed to assess their efficiency. To enhance transparency and reproducibility, the implementation of the proposed AVHH algorithm, along with the experimental configuration details, is available at https://github.com/tribhuvansingh88/AVHH-MATLAB-Code.