Ensemble Deep Learning for Forecasting Lassa Fever Outbreaks in Nigeria: Integrating Incidence and Weather Data for Early Warning

Lassa fever is one of the major public health challenges for many communities in Sub-Saharan West Africa. In the absence of an effective vaccine against Lassa fever infection, the development of forecasting models becomes crucial. Despite existing statistical and machine learning models, more advanced techniques are required to predict the outbreak of Lassa fever. The objective of this research is to develop a weighted ensemble deep learning model for forecasting Lassa fever outbreaks by incorporating weather factors into Lassa fever incidence data. Two Nigerian states (Ondo and Edo States), where the disease is more prevalent were used as case studies to validate the models. An ensemble of three deep learning techniques including, Long-Short Term Memory (LSTM), Bidirectional LSTM (BiLSTM), and Gated Recurrent Unit (GRU), was used to forecast Lassa fever cases 8 weeks and 12 weeks into the future. This is the first study to apply a weighted ensemble deep learning approach (LSTM-BiLSTM-GRU) for Lassa fever forecasting, addressing a critical gap in existing literature. Traditional models (e.g., ARIMA) fail to capture nonlinear weather-disease relationships, motivating our deep learning approach. Findings from the experiments show that the weighted ensemble model outperformed the individual deep learning models. This research has shown the potential of ensemble deep learning techniques in forecasting Lassa fever outbreaks, thereby enhancing epidemic preparedness and response from relevant stakeholders.