Implementation of DBSCAN and K-MEANS++ Methods for Flood Vulnerability Cluster Mapping in East Java Province, 2024

Flood disasters remain one of the most frequent natural hazards in Indonesia, particularly in East Java Province, where variations in rainfall, population density, and land use contribute to differing levels of flood vulnerability. Although numerous studies have explored flood susceptibility using geospatial and statistical models, comparative analyses of clustering algorithms specifically designed for complex regional topographies remain limited. This study aims to implement and compare two clustering algorithms Density Based Spatial Clustering of Applications with Noise (DBSCAN) and K-Means++ to map flood vulnerability patterns across the region. The urgency of this study lies in the province’s increasing flood frequency and its impact on infrastructure and livelihoods, which demand an adaptive and data driven spatial analysis approach. DBSCAN was selected for its ability to detect irregular, non-linear cluster shapes, while K-Means++ offers efficiency and stability through improved centroid initialization. The dataset includes hydrometeorological and socio-environmental indicators such as rainfall, elevation, slope, land cover, and population density. Clustering performance was assessed using the Silhouette Index (SI) and Davies Bouldin Index (DBI). Results showed that DBSCAN achieved a higher SI (0.3266) compared to K-Means++ (0.2453), indicating better cohesion and separation. Spatially, DBSCAN generated four distinct clusters corresponding to actual flood prone areas, particularly in Jember, Lumajang, Pasuruan, and Sidoarjo. These findings suggest that density-based clustering provides a more reliable representation of heterogeneous spatial flood patterns, supporting local governments in targeted mitigation planning and regional disaster risk management.