Analisis Pola Penyebaran Penyakit Campak Di Kabupaten Sumenep Dengan Algoritma Random Forest Berbasis Geospasial

Abstract

Measles remains a Highly contagious disease and poses a surveillance challenge in Sumenep Regency, whose administrative area consists of mainland and island subdistricts. This study aimed to analyze the spatial distribution of measles, forecast monthly cases, and implement the results in a WebGIS dashboard. The study used measles records from 2022–2025 obtained from the Sumenep Regency Health Office and subdistrict boundaries in GeoJSON format. Annual records for 2022–2024 were temporally disaggregated using the monthly distribution pattern of the detailed 2025 data, while individual 2025 records were aggregated by subdistrict and month. The resulting panel comprised 27 subdistricts, 48 monthly periods, and 1,296 observations. Spatial analysis compared Queen Contiguity and K-Nearest Neighbor weights, followed by Global Moran’s I and Local Indicators of Spatial Association with 999 permutations. Random Forest Regression used Lag-1 to Lag-4, month, and year, with a chronological split and five-fold expanding-window RandomizedSearchCV. KNN with k=5 produced a connected spatial network and significant positive autocorrelation (Moran’s I=0.3384; p=0.004). The tuned model achieved MAE=3.6021, RMSE=5.6325, and R²=0.3025, outperforming the Lag-1 baseline. Lag-1 was the dominant predictor. The integrated WebGIS presents case maps, LISA clusters, forecasts, and operational priority areas. The approach supports data-driven surveillance, although predictions remain moderate and tend to underestimate extreme increases.