Forecasting Lassa Fever Outbreak Progression with Machine Learning

Lassa fever (LF) is a zoonotic disease in West Africa. The disease is prone to epidemic tendencies, as cases have been recorded in non-endemic countries such as the UK, the USA, and Germany. Prompt surveillance of LF outbreaks and forecasting of future cases are vital for healthcare organizations. This study explored decision tree (DT) and k-nearest neighbor (kNN) architectures with time-dependent variables for model performance and reliable error estimates of weekly confirmed LF cases. LF models were benchmarked using the seasonal autoregressive integrated moving average (SARIMA). Results from SARIMA, DT, kNN, grid search of DT, and grid search kNN show that implemented models produced robust performance regardless of the LF confirmed case data complexity. Evaluation from the test set shows that kNN with grid search prediction outperformed other models except in kNN model, where the MAE had a lower error score, and the DT with grid search for MAPE had a lower error when compared to other models.