Analisis Pola Pembayaran Retribusi Daerah Menggunakan DBSCAN untuk Meningkatkan Efektivitas Penagihan

Abstract

This study analyzes regional levy payment patterns using Density-Based Spatial Clustering of Applications with Noise (DBSCAN) to support data-driven collection strategies at BPKPD Simalungun Regency. The main problem is that levy payment data have not been optimally used to distinguish compliant payers, risky payers, and unusual payment behavior. This research applies a descriptive exploratory quantitative data mining approach using 2,000 records with 15 columns. The main DBSCAN model focuses on five aggregate features: payment frequency, settlement ratio, log-transformed arrears, late-payment proportion, and average days late. The analytical stages include data collection, data cleaning, analysis dataset formation, feature standardization, DBSCAN modeling, parameter testing, PCA-based visualization, cluster evaluation, and interpretation of collection recommendations. The selected configuration uses eps = 0.5 and min_samples = 5. The results show that DBSCAN produces two clusters and eight noise records, or 0.40% of the total data. The silhouette score reaches 0.5737, indicating that the non-noise clusters are sufficiently separated for interpretation. Cluster A represents relatively compliant payers, Cluster B represents high-risk payers with lower settlement ratios and dominant late-payment behavior, while noise records indicate unusual payment patterns requiring verification. The findings show that DBSCAN can support early segmentation of levy payers, outlier detection, and more targeted collection prioritization for local revenue management.