<![CDATA[Understanding and predicting disease distribution patterns in primary healthcare settings require models capable of integrating both spatial and temporal dimensions. Traditional statistical approaches often fail to capture complex non-linear relationships across locations and time, leading to delayed detection of disease clusters. Objective: This study aims to develop a spatiotemporal machine learning framework to identify and forecast potential disease hotspots using electronic primary care records from 2024. Methods: The dataset comprised 5,343 patient visit records containing temporal, geographic (village-level), and clinical attributes. Data preprocessing included temporal aggregation, spatial encoding, and feature normalization. Three models—Gradient Boosting Machine (GBM), Temporal Random Forest (TRF), and Multi-EigenSpot—were trained and evaluated. Model performance was assessed using AUC, F1-score, and spatial accuracy metrics to ensure both predictive precision and spatial coherence. Results: Analysis revealed a clear seasonal pattern, with disease incidence peaking between April and August. Spatial mapping identified consistent hotspots in Sungai Asam and Beringin, accounting for over 70% of total cases. Among all tested models, Multi-EigenSpot achieved the best performance (AUC = 0.91; F1 = 0.86), effectively capturing multi-cluster spatial variability across months. Conclusions & Implications: The findings demonstrate that spatiotemporal learning models can significantly enhance disease surveillance and early warning capabilities in primary healthcare systems. Integrating spatial intelligence with explainable machine learning improves predictive accuracy, supports evidence-based policy, and enables targeted interventions for emerging disease hotspots in resource-limited settings.]]>
Copyrights © 2026
