Abstract
About this talk
Introduction: The COVID-19 pandemic has underscored the need for accurate predictive models to inform public health interventions. Traditional models like Autoregressive Integrated Moving Average (ARIMA) model assume a linear relationship and constant error variance over time with in COVID-19 case trends And Long Short-Term Memory (LSTM) networks exceed in imprison nonlinear dependencies but struggles with high-dimensional data, resulting in inaccurate predictions and poor fit due to high computational demands. This study aims to robust in accuracy of COVID-19 case predictions in Malaysia by developing a hybrid ARIMALSTM model that combines the strengths of both ARIMA and LSTM approaches.
Methodology: We trained and validated the hybrid model using real-time COVID-19 case data from the Malaysian Ministry of Health, focusing on September 12 to September 18, 2021. The best hyperparameter of the proposed model for ARIMA p = 9, d = 2, q = 2 and LSTM epochs = 472, batch verbose = 1 respectively. The model's execution was analogous to standalone ARIMA and LSTM models using key metrics: Mean Squared Error (MSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE), and Relative Root Mean Square Error (RRMSE).
Results: Our analysis revealed the ARIMA-LSTM hybrid model performed much better across all criteria than either of the solo models. The lowest values of error metrics in ARIMA -LSTM hybrid model compared to others standalone models ARMA and LSTM. The values of ARIMA-LSTM hybrid model are MSE = 0.02, MAE = 0.10, MAPE = 2.80, RMSE = 0.14 and RRMSE = 0.05, respectively.
Conclusion: The ARIMA-LSTM hybrid model offers a spirited tool for predicting COVID-19 case trends, providing valuable insights for policymakers. This study demonstrated the prospect of mixed modelsin epidemiological forecasting, with broader implications for managing public health crises
Speaker
About the speaker
