Binning-Based Silhouette Approach to Find the Optimal Cluster Using K-Means

Abstract

Clustering is one of the critical parts of machine learning algorithms. K-Means clustering is the standard technique that various data analysts use for clustering the data among the various clusters. Even though the K means clustering algorithm can work effectively, there is a need to tune the value of K according to the dataset under consideration. The process of tuning for the value of k requires the execution of the Kmeans algorithm with different values of k. The values of k with the best cluster quality based on specific metrics are selected. The elbow method and silhouette coefficient is the most popular approach for selecting the number of clusters. However, both approaches are time-consuming, as they have to execute K-means for each value of k to find a good score. This approach is iterative. This paper proposed a divide and conquer approach that performs the same task in less time. From the proposed approach, the task has been completely 2.5 times faster in comparison to the iterative method at the cost of some memory required to store the results.

Does diversity facilitate consumer decisions: a comparative perspective based on single-category versus multi-category products

“…Study 2 examines the moderating effect of heterogeneous visit types. The number of clusters is determined using the silhouette coefficient method (Punhani et al, 2022). The calculation of the silhouette coefficient is as follows:…”

Enhanced Firefly-K-Means Clustering with Adaptive Mutation and Central Limit Theorem for Automatic Clustering of High-Dimensional Datasets

“…It has the unique capability of automatic subdivision, which gives it the advantage of handling multimodal optimization problems compared with other metaheuristic algorithms [28]. It records superior performance in clustering analysis due to the clustering process's high non-linearity and sub-optimal distraction [6]. A systematic review of the FA is presented in [28] describing the various characteristics and variants of the algorithm.…”

“…Two variants of the FA were hybridized with K-means by [6] to resolve the K-means algorithm's local optimal trap and initialization sensitivity problems. Their hybrid model was based on the FA's unique property of automatic sub-division and ability to tackle multimodal optimization problems of the firefly algorithm.…”

“…It employs a distance-based optimization criterion to minimize the intra-cluster distance and maximize the inter-cluster distance. Several limitations, however, have been identified in using the K-means clustering algorithm, including sensitivity to initialization parameters, undesirable sample distribution vulnerability, and susceptibility to noise [6,7]. The optimum performance of K-means is premised on the specification of the correct number of clusters (which is difficult to determine in a real-life dataset) and the selection of the optimum number of the initial cluster centroid.…”

“…In recent times, hybridizing K-means with metaheuristics has also assisted in solving most of the problems associated with real-life datasets based on their global search capability for the optimized number of clusters and cluster centroids [6,[13][14][15][16][17][18][19][20][21]. Primarily, metaheuristics algorithms have been used to help the K-means algorithm to escape from local optima convergence by searching for global optimum initial cluster centroids.…”