Unveiling Banking Customer Pattern Through K-Means and DBSCAN Cluster Evaluation

The banking sector’s Customer segmentation plays a crucial role in understanding diverse customer behaviors and designing targeted financial strategies. This study employs K-Means and DBSCAN algorithms to analyze banking data and uncover customer behavior patterns. The clustering process with K-Means (k=5) produced five distinct groups of customers, each characterized by differences in loan amount, credit limit, account balance, and rewards points. These clusters demonstrate meaningful differentiation, such as customers with high loan exposure and moderate credit limits compared to groups with smaller loans but higher rewards accumulation. In contrast, DBSCAN (eps=1.9, min_sample=5) produced three clusters. Most data points were concentrated in one large cluster, while a few formed small groups with considerable noise. Evaluation metrics further confirmed the superiority of K-Means over DBSCAN. K-Means achieved better performance (Silhouette = 0.09, DBI = 2.8, CHI =820), indicating moderate separation and interpretability. In contrast, DBSCAN showed a negative Silhouette Score (-0.354) and a low Calinski-Harabasz Index (3.608), indicating poorly defined clusters. These results suggest that K-Means is more effective for banking customer segmentation, providing clearer profiling insights, whereas DBSCAN is less suitable due to the dataset’s homogeneity and distribution.