The article presents a comprehensive approach to optimizing clustering algorithms within decision support systems (DSS) in big data environments. It analyzes issues of scalability, computational complexity, and result stability that are typical of classical methods such as K-Means, DBSCAN, and Agglomerative Clustering. An improved hybrid algorithm, K-Means++ Hybrid, is proposed, combining parallel computing mechanisms, adaptive parameter tuning, and dynamic control of the iterative search process. The methodological foundation of the research is based on systems analysis, mathematical modeling, and experimental testing using datasets from the UCI Repository and GPU acceleration technologies (CUDA). Experimental results confirm that the proposed approach reduces clustering execution time by approximately 43% compared to baseline algorithms, while increasing the silhouette coefficient to 0.73 and reducing CPU energy consumption by 20–25%. The resulting model demonstrates high robustness when processing heterogeneous datasets and can be integrated into systems for traffic flow analysis, financial risk assessment, and environmental monitoring. The developed approach provides a foundation for building adaptive intelligent data analysis modules that support scalability, result interpretability, and real-time operation in streaming analytics systems. Future research should focus on integrating hybrid clustering with deep learning models and evolutionary optimization algorithms.
Ganna et al. (Sun,) studied this question.