International Journal of Advanced Research in Computer and Communication Engineering

Abstract: Malaria remains a significant public health challenge in Nigeria, with Bayelsa State experiencing persistent high transmission rates despite control efforts. This study developed a comprehensive deep learning-based malaria forecasting and early warning system for the eight Local Government Areas (LGAs) in Bayelsa State. The system utilizes Long Short-Term Memory (LSTM) neural networks enhanced with Principal Component Analysis (PCA) to predict malaria cases and generate early warnings through 2028. Historical malaria surveillance data from 2019-2024 was integrated with environmental variables including rainfall, temperature, humidity, and vector density indices. The model incorporates sophisticated feature engineering, including lag variables, seasonal indicators, and intervention coverage metrics to capture complex temporal patterns. PCA dimensionality reduction improved computational efficiency by 37% while enhancing predictive accuracy. The LSTM+PCA model achieved exceptional performance with R² = 0.939, RMSE = 14.04, and MAE = 10.02, substantially outperforming traditional approaches including ARIMA (R² = 0.849) and baseline models. Early warning thresholds were established using percentile-based methods, with LGA-specific values ranging from 146.5 to 179.5 cases, enabling localized outbreak detection. Model interpretability was enhanced through SHAP (SHapley Additive exPlanations), permutation importance, and Partial Dependence Plot (PDP) analyses, revealing climate variables and lagged malaria cases as primary transmission drivers. The system provides forecasts extending to 24 months, though accuracy assessment was limited to the test period, demonstrating sustained low-risk classifications across all LGAs through 2028. This innovative approach offers a robust tool for public health authorities to implement targeted, data-driven malaria control strategies, with real-time prediction capabilities under 9 milliseconds enabling integration into existing health information systems for improved epidemic preparedness and response.

Keywords: Malaria forecasting, LSTM neural networks, Early warning system, Bayelsa State, Nigeria.

Downloads: | DOI: 10.17148/IJARCCE.2025.14806