Hybrid Prediction of Daily COVID-19 Infections Using Machine Learning and SIR Model Integration

The COVID-19 pandemic has highlighted significant challenges in predicting the daily infectious individuals, a critical aspect for managing public health responses. Accurate prediction models are essential, but epidemic models, alongside traditional statistical and machine learning models require extensive data, which may have accurate predictions but may not be timely enough to handle a pandemic immediately in real-life scenarios. This paper explores a hybrid approach, utilizing machine learning to estimate the basic reproduction number (Ro) and integrating it into the SIR (Susceptible-Infected-Recovered) model for more reliable predictions during the early stages of a pandemic using limited data. This study evaluates the effectiveness of this approach using historical COVID-19 data from Thailand and compares it to existing methods. The proposed hybrid model is evaluated against traditional SIR models with fixed Ro and various machine learning models, including Random Forest Regressor, RNN, LSTM, and GRU. Prediction accuracy is measured by mean absolute error (MAE). Our approach demonstrates superior performance in predicting daily infectious individuals during the first wave of COVID-19 in Thailand, even with limited early-stage data, and effectively adapts to changing epidemic dynamics influenced by policy responses. Results indicate that the hybrid model significantly outperforms both traditional SIR models and other machine learning approaches, showcasing its robustness and adaptability in real-world pandemic management scenarios.