Improving Seasonal Influenza Forecasting Using Time Series Machine Learning Techniques

Abstract

Influenza is a highly contagious respiratory disease and is still a serious threat to public health all over the world. Forecasting techniques help in monitoring seasonal influenza and other influenza-like diseases and also in managing resources appropriately to formulate vaccination strategies and choose appropriate public health measures to reduce the impact of the disease. The aim of this investigation is to forecast the monthly incidence of seasonal flu in Saudi Arabia for the years 2020 and 2021 using the XGBoost model and compare it with ARIMA and SARIMA models. The results show that the XGBoost model has the lowest values MAE, MAE, and RMSE compared to the ARIMA and SARIMA models and the highest value of R-squared (R2). This study compares the accuracy of the XGBoost model with ARIMA and SARIMA models in providing a forecast of the number of monthly seasonal influenza cases. These results confirm the notion that the XGBoost model has a higher accuracy of prediction than that of the ARIMA and SARIMA models, mainly due to its capacity to capture complex nonlinear relationships. Therefore, the XGBoost model could predict monthly occurrences of seasonal influenza cases in Saudi Arabia.