Time-Series Forecasting of Lassa fever Outbreaks in Nigeria using Facebook Prophet; Application of Machine Learning in Early Warning, Alert and Response Systems
Franco Apiyanteide*, Adenonmon Monday Osagie, Chaku Shammah Emmanuel, Abubakar Muhammad Auwal and Justus Ajaegbu Ezechinyere
ABSTRACT
Since the first cases of Lassa fever were diagnosed in Northeastern Nigeria in 1969, the disease has been one of the neglected tropical diseases with no definitive cure, no approved vaccines and poor public health measures needed to mitigate the burden of the disease in the country. This study aimed to provide time series forecasting of Lassa fever in Nigeria using Facebook Prophet, to improve the understanding of the seasonal patterns and trends of the disease in Nigeria. Data for the study were obtained from the Nigeria Centre for Disease Control and Prevention (NCDC) weekly situation report website from 2018 to 2025, using a Microsoft Excel template that captured relevant information on the disease (year, epidemiological week, number of cases, and number of deaths) from the agency's situation report. Data collected was processed, split into training and testing datasets that were used for forecasting and performance evaluation of the model. The model was fit and used to make a 5-year forecast using the Facebook Prophet Library in Python version 3.13 to obtain a 5-year forecast for the disease in Nigeria from 2025 to 2030, as well as gain an in-depth understanding of the seasonality and trend of the disease. Our result showed that a total of 7728 Lassa fever confirmed cases and 1288 confirmed deaths, with a Case Fatality Rate (CFR) of 16.7% were recorded in Nigeria from 2018 to 2025. There was an upward trend for the number of cases and deaths from Lassa fever over the study period, with January to March, which corresponds to the dry season in the country, being the peak for the disease. Performance evaluation was conducted for confirmed cases of the disease using Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Coefficient of Determination (R2) and Mean Poisson Deviance (D) and this showed 9.44, 12.51, 0.57 and 6.3 respectively. Using Facebook Prophet, by 2030, there will be an estimated 9100 (95% Confidence Interval (CI): 1937 – 16268) confirmed cases and 1721 (95% CI: 313 – 3125) confirmed deaths from Lassa fever in Nigeria. There is therefore an urgent need to apply machine learning algorithms in understanding infectious disease trend and seasonality which could guide early warning, alert and response systems which is proactive rather than the traditional reactive response to outbreaks when several morbidity and mortality have already occurred.
