Forecasting COVID-19 cases using time series modeling and association rule mining

Area of Science:

• Epidemiology and Public Health
• Data Science and Machine Learning
• Health Services Research

Background:

• COVID-19 pandemic necessitates robust crisis management strategies.
• Accurate prediction of case numbers is crucial for resource allocation.
• Identifying prognostic factors aids in understanding disease progression and patient outcomes.

Purpose of the Study:

• To evaluate the combined effectiveness of Autoregressive Integrated Moving Average (ARIMA) and Association Rule Mining (ARM) for COVID-19 case prediction.
• To identify key prognostic factors influencing COVID-19 patient outcomes and hospital stays.
• To develop an optimized model for efficient COVID-19 crisis management.

Main Methods:

• Analysis of 3685 COVID-19 patient data from Thailand (March 2020 - August 2021).
• Application of Autoregressive Integrated Moving Average (ARIMA) and its derivative Autoregressive Integrated Moving Average with Exogenous Variables (ARIMAX) models.
• Utilized Association Rule Mining (ARM) to identify significant prognostic factors from patient data.

Main Results:

• The ARIMA (2, 2, 2) model predicted COVID-19 cases with R²=0.5695, RMSE=29.7605, MAE=27.5102.
• The ARIMAX (1, 1, 1) model, incorporating ARM-identified factors (symptoms, age, underlying diseases), showed improved prediction (R²=0.5695, RMSE=27.7508, MAE=23.4642).
• Longer hospital stays (>14 days) linked to healthcare worker status and underlying diseases. Worsening cases associated with symptoms, pregnancy, metabolic syndrome, and age >65.

Conclusions:

• The ARIMAX model, enhanced by ARM-identified prognostic factors, shows significant potential for predicting COVID-19 case numbers.
• This predictive capability can optimize hospital resource preparation and management during pandemics.
• The study highlights the value of integrating time series analysis with data mining for public health crisis response.