Systematic Literature Review: A Comparison of Clustering Methods in Data Mining

Abstract

Clustering is one of the fundamental techniques in data mining used to group data instances based on inherent similarities without relying on predefined labels. This technique plays a crucial role in numerous domains, including customer behavior analysis, pattern recognition, anomaly detection, bioinformatics, and many other applications that require a deeper understanding of hidden structures within data. Over the past decades, a wide range of clustering methods has been developed such as K-Means, DBSCAN, Hierarchical Clustering, density-based approaches, model-based clustering, and more recent algorithms that incorporate machine learning and deep learning paradigms. Each method offers distinct advantages and limitations and is suited for different data characteristics and analytical objectives. The SLR process includes identifying relevant articles, screening for quality and eligibility, extracting essential data, and synthesizing findings according to predefined systematic criteria. The primary aim of this review is to identify emerging research trends, understand methodological advancements, assess the performance of different clustering methods across diverse data contexts such as varying dataset sizes, noise levels, dimensionality, and cluster distributions and provide insights into the key factors that influence the selection of appropriate clustering techniques. The findings of this review indicate that no single clustering method consistently outperforms others in all scenarios. Certain algorithms may produce optimal results for low-dimensional datasets yet perform inadequately when applied to complex, high-dimensional data. Conversely, some methods are effective at identifying clusters with irregular shapes but require sensitive parameter tuning or exhibit higher computational costs. Therefore, the choice of clustering technique should be guided by the specific characteristics of the dataset, the objectives of the analysis, and evaluation criteria such as accuracy, computational efficiency, interpretability, and robustness to noise. Overall, this review aims to serve as a comprehensive reference for researchers, practitioners, and decision-makers in selecting the most suitable clustering method for their specific analytical needs. Additionally, the study highlights potential avenues for future research, including the development of hybrid algorithms, automated parameter selection techniques, and the integration of clustering with modern machine learning approaches to enhance performance and adaptability across various data environments