Leveraging Random Forest (RF) and Long Short-Term Memory Algorithms (LSTM) for Enhanced Cholera Outbreak Prediction and Response System in Zambia

Zambia has experienced recurrent cholera outbreaks since 1977, with the 2023-2024 epidemic recording over 10,887 cases and 432 deaths, the most severe in the nation's history. These outbreaks persist due to inadequate surveillance, delayed reporting, and a lack of predictive capacity regarding environmental triggers (like rain and temperature), resulting in slow detection and a delayed response. This paper proposes a hybrid AI-driven Early Warning System (EWS) that integrates the Random Forest (RF) for case classification and Long Short-Term Memory (LSTM) networks for temporal outbreak forecasting. The system combines community-based mobile reporting with environmental data analysis to facilitate real-time surveillance. Using a proxy dataset (YEM-CHOLERA-EOC-DIS-WEEK-20160424-20200621.csv) to simulate the outbreak dynamics in the absence of local digital records, the proposed RF-LSTM ensemble model has demonstrated superior performance compared to the standalone models. The hybrid model has achieved a classification F1-score of 86% and a forecasting Root Mean Squared Error (RMSE) of 0.134, significantly outperforming the individual RF and LSTM models. This study presents a scalable, proactive framework for mitigating future cholera epidemics in resource-constrained settings.