Forecasting COVID-19 pandemic in Alberta, Canada using modified ARIMA models.

Area of Science:

• Time Series Analysis
• Epidemiological Forecasting
• Statistical Modeling

Background:

• Auto Regressive Integrated Moving Average (ARIMA) models are widely used for time series forecasting.
• Standard ARIMA models assume constant variance, which is often violated in real-world data, leading to inaccurate confidence intervals.
• Heteroscedasticity, where variance changes over time, poses a challenge for traditional ARIMA models.

Purpose of the Study:

• To develop and validate a revised ARIMA model capable of handling time series with heteroscedasticity.
• To improve the accuracy of confidence intervals in time series forecasting when variance is non-constant.
• To apply the revised model for forecasting COVID-19 incidence in Alberta, Canada.

Main Methods:

• Constructed multiple historical ARIMA models using publicly available COVID-19 data from Alberta, Canada.
• Applied time series analysis across different periods to capture evolving patterns.
• Modified forecasted values from general ARIMA models by incorporating calculated differences between forecasted and actual values, including confidence intervals.

Main Results:

• The proposed revised ARIMA method yielded lower average forecasted COVID-19 incident cases compared to the general ARIMA model.
• The 95% confidence intervals for incidence forecasts were narrower with the revised method, indicating increased precision.
• Forecasts predicted an initial increase in average incidence followed by an exponential decrease over the 13-week period.

Conclusions:

• The revised ARIMA method effectively addresses heteroscedasticity and automates ARIMA model selection for improved forecasting.
• The model accurately forecasts longer-term trends, suggesting a potential decrease but not elimination of COVID-19 incidence in the next 13 weeks.
• Persistent public health interventions, guided by accurate forecasts, are crucial for eventual transmission control.