Perbandingan Metode Clustering pada Segmentasi Konsumen Berbasis Data Profil Pelanggan

Abstract

Customer segmentation is an important aspect of customer data analysis to understand customer characteristics and support data-driven decision making. However, limitations in transactional data availability encourage the use of non-transactional customer profile data as an alternative for segmentation. This study aims to compare the performance of several clustering methods in customer segmentation based on customer profile data. This research employs a quantitative approach using data mining techniques. The dataset used is the Customers Purchase Behavior dataset from Kaggle, consisting of 72,637 customer records. The analyzed variables include lifestage and premium customer. Categorical data were transformed through encoding and normalized prior to the application of three clustering methods: K-Means Clustering, Hierarchical Clustering, and DBSCAN. Cluster quality was evaluated using the Silhouette Coefficient, Davies–Bouldin Index, and Calinski–Harabasz Index. The results indicate that K-Means Clustering achieved the best cluster quality, demonstrated by the highest Silhouette score, the lowest Davies–Bouldin Index, and the highest Calinski–Harabasz Index compared to the other methods. Hierarchical Clustering showed moderate performance, while DBSCAN was less effective due to the relatively homogeneous characteristics of the customer profile data. These findings suggest that the effectiveness of clustering methods is highly dependent on data characteristics. K-Means Clustering is recommended as the most suitable method for customer segmentation based on non-transactional customer profile data in this study.