EVDHM-ARIMA-Based Time Series Forecasting Model and Its Application for COVID-19 Cases

Area of Science:

• Time-series analysis
• Statistical modeling
• Epidemiological forecasting

Background:

• Accurate time-series forecasting is crucial for effective decision-making.
• Nonstationary time series present significant challenges for traditional forecasting models.
• Existing methods may struggle with the complexity and dynamic nature of real-world data.

Purpose of the Study:

• To develop a robust forecasting model for nonstationary time series.
• To enhance prediction accuracy by addressing data nonstationarity.
• To apply the model for forecasting COVID-19 new daily cases.

Main Methods:

• Utilized eigenvalue decomposition of Hankel matrix (EVDHM) to decompose and reduce nonstationarity in time series.
• Employed autoregressive integrated moving average (ARIMA) models for forecasting subcomponents.
• Optimized ARIMA parameters using a genetic algorithm (GA) to minimize Akaike information criterion (AIC).
• Applied the Phillips-Perron test (PPT) to identify time-series nonstationarity.

Main Results:

• The EVDHM-ARIMA model demonstrated high efficacy in forecasting.
• Successfully decomposed nonstationary time series into manageable subcomponents.
• Accurate predictions of daily new COVID-19 cases were achieved for India, USA, and Brazil.
• The genetic algorithm effectively optimized ARIMA parameters for improved accuracy.

Conclusions:

• The proposed EVDHM-ARIMA method offers a powerful approach for nonstationary time-series forecasting.
• This technique provides reliable predictions for critical applications like epidemiological surveillance.
• The study validates the model's effectiveness on real-world pandemic data.