Improvement of Time Forecasting Models Using Machine Learning for Future Pandemic Applications Based on COVID-19 Data...

Area of Science:

• Computational epidemiology
• Time-series forecasting
• Machine learning applications in public health

Background:

• Accurate forecasting of Coronavirus Disease-19 (COVID-19) is crucial for public health response.
• Existing linear models like autoregressive integrated moving average (ARIMA) struggle with the complex linear and non-linear patterns in COVID-19 data.
• This limitation leads to imprecise and inefficient predictions, hindering effective monitoring and prevention strategies.

Purpose of the Study:

• To propose novel intelligence-based prediction methods for enhanced COVID-19 forecasting.
• To develop and evaluate a hybrid model, autoregressive integrated moving average-least-squares support vector machine (ARIMA-LSSVM), for improved accuracy and efficiency.
• To compare the performance of the proposed ARIMA-LSSVM model against ARIMA, support vector machine (SVM), and least-squares support vector machine (LSSVM) models.

Main Methods:

• Implementation of a hybrid forecasting approach integrating ARIMA with LSSVM.
• Utilized real-world COVID-19 datasets: daily new cases, daily new deaths, and daily new recovered cases.
• Performance evaluation using statistical measures: Mean Square Error (MSE), Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE).

Main Results:

• The proposed ARIMA-LSSVM model demonstrated superior performance across all tested COVID-19 datasets (cases, deaths, recoveries).
• Empirical results from Malaysian datasets showed significantly lower MSE, RMSE, MAE, and MAPE values for both training and testing phases compared to ARIMA, SVM, LSSVM, and ARIMA-SVM.
• The proposed model achieved a higher percentage of error reduction, indicating more accurate and efficient predictions closer to the true values.

Conclusions:

• The hybrid ARIMA-LSSVM model significantly enhances the accuracy and efficiency of COVID-19 time-series forecasting.
• This intelligent approach provides more reliable predictions than traditional and other machine learning models.
• The proposed model represents a promising tool for improving future pandemic response and management strategies.