EXPLAINABLE MODELS FOR EARLY PREDICTION OF INFECTIOUS DISEASE OUTBREAKS: INTEGRATING SPATIO-CLIMATIC AND SECURITY INDUCED MIGRATION DATA IN NORTH-EASTERN NIGERIA.

Abstract

North-Eastern Nigeria continues to grapple with the dual challenge of infectious diseases such as malaria, cholera, measles, and scabies, compounded by persistent conflict-driven displacement. Existing surveillance systems remain limited, often undermined by underreporting, delayed responses, and inadequate integration of ecological and socio-political drivers. This study-in-progress proposes an explainable artificial intelligence (XAI) framework for early outbreak prediction that integrates spatio-climatic variables with conflict-induced migration data. Spatio-temporal grids (2010–2025) are being constructed to derive lagged indicators that capture delayed impacts of weather and displacement on transmission dynamics. The methodological approach combines ensemble learners (Random Forest, XGBoost) with deep spatio-temporal networks (LSTM, ConvLSTM), while outbreak definitions are based on 95th percentile anomalies. Explainability mechanisms, including SHAP, saliency maps, counterfactual analysis, and expert validation, are incorporated to ensure interpretability. Although analysis is ongoing, the study aims to benchmark performance against climate-only, migration-only, and traditional baselines. The anticipated outcome is a transparent, reliable predictive framework capable of informing epidemic preparedness and intervention planning in fragile and conflict-affected contexts.