Optimizing Fuzzy C Means Clustering Algorithm: Challenges and Applications

Abstract

The Fuzzy C-Means clustering technique is one of the most popular soft clustering algorithms in the field of data segmentation. However, its high time complexity makes it computationally expensive, when implemented on very large datasets. Kolen and Hutcheson [1] proposed a modification of the FCM Algorithm, which dramatically reduces the runtime of their algorithm, making it linear with respect to the number of clusters, as opposed to the original algorithm which was quadratic with respect to the number of clusters. This paper proposes further modification of the algorithm by Kolen et. al., by suggesting effective seed initialisation (by Fuzzy CMeans++, proposed by Stetco et. al. [2]) before feeding the initial cluster centers to the algorithm. The resultant model converges even faster. Empirical findings are illustrated using synthetic and real-world datasets. Finally, we check the algorithm’s robustness to perturbations in the data.