Perbandingan Metode Clustering K-Means, GMM, dan DBSCAN Berbasis Fitur RFM

Abstract

In the digital banking era, understanding customer behavior has become essential for delivering relevant services and maintaining competitiveness. This study aims to develop and evaluate customer segmentation models by leveraging an extended RFM (Recency, Frequency, Monetary) model, incorporating both behavioral and demographic attributes. Preprocessing, feature engineering, handling outliers, and standardization were done on the data using a dataset of 100,000 bank transaction records from Kaggle. DBSCAN, Gaussian Mixture Model (GMM), and K-Means were the three clustering techniques that were employed and contrasted. The clustering performance was evaluated using the Silhouette Score, Calinski-Harabasz Index (CHI), and Davies-Bouldin Index (DBI). The output of DBSCAN was too noisy to be useful in the business world, despite having the highest scores on Silhouette: 0,667 and lowest score on DBI: 0,396. K-Means offered the most interpretable segmentation with five ideal clusters (Silhouette: 0,308; DBI: 0,957; CHI: 6191), identifying customer groups ranging from highly active to potentially inactive. The findings highlight the synergy between transactional features and clustering algorithms in generating actionable insights for banking strategy.