Comparison of Elbow and Silhouette Methods in Optimizing K-Prototype Clustering for Customer Transactions

Abstract

This research presents a comparative analysis of the Elbow and Silhouette methods to identify the ideal number of clusters in applying the K-Prototypes algorithm for customer grouping using purchase transaction data. The K-Prototypes algorithm is employed due to its ability to handle both numerical and categorical data simultaneously. Customer purchase transaction data from the Point of Sale (POS) system is analyzed through preprocessing, feature transformation, and attribute segmentation stages before being clustered using the K-Prototypes algorithm. To identify the optimal cluster count, this study employs two methods: the Elbow and the Silhouette method. The results indicate that the Elbow method produces 2 clusters with a model evaluation score of 0.6368, while the Silhouette method suggests 2 clusters with a slightly lower score of 0.6186. In terms of computational efficiency, the Elbow method also demonstrates a faster processing time results highlight the significance of choosing an appropriate method for identifying the ideal number of clusters, ensuring it aligns with the specific goals of the analysis, whether emphasizing superior inter-cluster distinction or favoring a more parsimonious model configuration.