<![CDATA[West Java Province exhibits high disaster vulnerability, necessitating accurate risk zone mapping for mitigation purposes. This study aims to conduct a comparative performance analysis of K-Means and DBSCAN algorithms for clustering disaster-prone areas. The comparison is important because K-Means, as a centroid-based algorithm, is sensitive to outliers, whereas DBSCAN, as a density-based method, is theoretically more suitable for complex disaster-risk data and capable of identifying anomalies. A quantitative approach was applied to secondary data on disaster incidents (Floods, Landslides, Earthquakes, Fires, and Extreme Weather) for the 2015-2024 period across 27 regencies/cities. Following normalization using StandardScaler, model performance was evaluated through Silhouette Score (SI) comparison and visual analysis of the formed cluster structures. The results reveal a paradoxical finding: although K-Means (K=2) numerically outperformed DBSCAN (0.468) with an average SI of 0.502, it demonstrated internal validity failure due to negative silhouette scores indicating misclassification in extreme regions. Conversely, DBSCAN proved superior in representing the natural data structure with its capability to isolate 6 anomalous regions as noise. Further temporal sensitivity analysis revealed significant risk dynamics, where the number of noise regions increased drastically from 5 regions in the 2015-2019 period to 10 regions in the 2020-2024 period. This indicates that disaster patterns in West Java are becoming increasingly irregular and unpredictable, positioning DBSCAN as the recommended robust method for complex risk mapping.]]>
