PERFORMANCE ANALYSIS OF K-MEANS AND FCM ON SHAPE AND DENSITY VARYING CLUSTERS

Abstract

This study conducts a comprehensive performance analysis of K-means and Fuzzy C means (FCM) clustering algorithms based on various distance metrics, including Euclidean, Manhattan, Mahalanobis, Minkowski, and Cosine distances. Clustering algorithms are essential tools for organizing data into meaningful groups. K-means and FCM are widely used algorithms in this context, with K-means focusing on crisp partitions and FCM providing fuzzy partitions. By employing different distance metrics, we explore how these algorithms perform under diverse similarity measures, capturing various aspects of data dissimilarity. Through extensive experimentation on benchmark datasets, we evaluate the clustering quality and computational efficiency of K-means and FCM algorithms using each distance metric. The evaluation metrics include intra-cluster distance, inter-cluster distance, silhouette coefficient, and clustering stability. Additionally, we analyze the runtime performance of the algorithms to assess their computational efficiency and scalability. The results of our analysis provide valuable insights into the performance characteristics of K-means and FCM algorithms when applied with different distance metrics. We identify scenarios where one algorithm outperforms the other, shedding light on the suitability of each algorithm and distance metric combination for specific data characteristics and clustering objectives. This study contributes to the existing body of knowledge by offering a comprehensive comparison of K-means and FCM algorithms based on diverse distance metrics, including Euclidean, Manhattan, Mahalanobis, Minkowski, and Cosine distances. The findings of this analysis can guide researchers and practitioners in selecting the most suitable algorithm and distance metric combination for their clustering tasks, leading to improved clustering accuracy and efficiency.