Sanitation Risk Prediction Model Based on G.I.S and A.I to Improve the Residential Sanitation Index in Palembang

Abstract

Residential sanitation conditions vary spatially across Palembang City, but a transparent and reproducible framework is needed to convert heterogeneous sanitation data into area-level priorities. This ecological cross-sectional study assessed all 18 subdistricts of Palembang by integrating a composite sanitation-risk index with Geographic Information System (GIS) mapping and an exploratory Random Forest analysis. The composite index was constructed from four direct sanitation and environmental indicators: access to improved latrines, access to improved water, wastewater-management access (SPAL), and illegal waste-disposal points. Indicators were min-max normalized to a common 0-1 risk direction and aggregated using equal weights. Population density, building density, diarrhea rate, and dengue rate were included as contextual variables and were not used to construct the index. Risk scores ranged from 13.09 to 73.83. Kertapati had the highest score (73.83), followed by Seberang Ulu II (54.08) and Seberang Ulu I (47.85). Tertile classification identified six high-, six medium-, and six low-priority subdistricts. Random Forest regression was evaluated using Leave-One-Out Cross-Validation because only 18 ecological units were available. The model showed MAE=14.01, RMSE=17.05, and R²=-0.15, and did not outperform a mean-prediction baseline, indicating limited out-of-sample predictive performance in the present dataset. Accordingly, Random Forest results were interpreted as exploratory and not as causal evidence or as an externally validated operational prediction model. The combined GIS and composite-index framework provides a reproducible basis for relative spatial prioritization, while future predictive modeling requires independent sanitation-risk labels, finer spatial units, multi-year data, and external validation.