Comparative Evaluation of Predictive Models for Malaria Cases in Sierra Leone

Malaria remains a major public health challenge necessitating accurate predictive models to inform effective intervention strategies in Sierra Leone. This study compares the performance of Holt-Winters’ Exponential Smoothing, Harmonic, and Artificial Neural Network (ANN) models using data from January 2018 to December 2023, incorporating both historical case records from Sierra Leone’s Health Management Information System (HMIS) and meteorological variables including humidity, precipitation, and temperature. The ANN model demonstrated superior performance, achieving a Mean Absolute Percentage Error (MAPE) of 4.74% before including climatic variables. This was further reduced to 3.9% with the inclusion of climatic variables, outperforming traditional models like Holt-Winters and Harmonic, which yielded MAPEs of 22.53% and 17.90% respectively. The ANN’s success is attributed to its ability to capture complex, non-linear relationships in the data, particularly when enhanced with relevant climatic variables. Using the optimized ANN model, we forecasted malaria cases for the next 24 months, predicting a steady increase from January 2024 to December 2025, with seasonal peaks. This study underscores the potential of machine learning approaches, particularly ANNs, in epidemiological modelling and highlights the importance of integrating environmental factors into malaria prediction models, recommending the ANN model for informing more targeted and efficient malaria control strategies to improve public health outcomes in Sierra Leone and similar settings.