Abstract
Introduction: Clustering algorithms play a key role in grouping data objects based on their similarities. A popular method, K-means, works by repeatedly adjusting the center of each cluster until convergence is achieved. This method, especially in the PAM form, is widely used in clustering analysis for its effectiveness in separating data. Clustering, an unsupervised learning technique, is very effective in discovering hidden patterns within datasets. Clustering focuses on dividing data into meaningful groups, rather than using predefined labels, as supervised algorithms do. By finding underlying structures and connections, this technique helps gain a deeper understanding of complex data. Research significance: This research is of great importance for the study of clustering algorithms developed for sparse industrial datasets. It aims to provide useful insights and standards for improving clustering performance in industrial settings by examining and contrasting five main clustering approaches: partitioning, hierarchical, density, grid, and model-based methods. Methology: Some alternative suppliers include K-Means, Hierarchical Clustering, DBSCAN, Gaussian Mixture, Sample, Iterated-Bisection, Agglomerative, and Clustering. Evaluation criteria include Silhouette Score, Davis-Bouldin Index, KalinskyHarapas Index, Cluster Cohesion, Execution Time, Memory Usage, and Sensitivity to Noise. Result: According to the results, DBSCAN was ranked highest, while Clustering was ranked lowest. DBSCAN has the highest value for Clustering algorithms according to the WPM Method approach.