Classification of Hydrometeorological Disaster Vulnerability Across Indonesian Provinces Using the KNN Algorithm Based on 2024 Podes Data

Abstract

Hydrometeorological disasters have increasingly posed significant challenges to regional resilience in Indonesia, driven by climate variability and uneven mitigation capacity across provinces. This study aimed to classify hydrometeorological disaster vulnerability across all Indonesian provinces using a machine learning approach based on the 2024 Village Potential Statistics dataset. A supervised learning framework was implemented using the k-Nearest Neighbor algorithm to integrate physical exposure indicators, including riverbank and slope settlements as well as river proximity, with mitigation capacity variables such as Early Warning Systems and evacuation infrastructure. Provincial-level data were aggregated, normalized, and processed following the Knowledge Discovery in Databases methodology. The classification results categorized provinces into low, medium, and high vulnerability levels, revealing that mitigation capacity played a critical role in moderating disaster vulnerability beyond physical exposure alone. Model evaluation demonstrated strong performance, with a high discriminative capability and balanced accuracy across classes, indicating that the selected k-Nearest Neighbor configuration was suitable for heterogeneous socio-environmental data. The findings highlighted the importance of preparedness infrastructure in reducing disaster risk and provided a transparent, data-driven framework to support evidence-based disaster management and policy planning at the national scale.