Event-Specific Transmission Forecasting of SARS-CoV-2 in a Mixed-Mode Ventilated Office Room Using an ANN

Area of Science:

• Environmental Science
• Epidemiology
• Artificial Intelligence

Background:

• Emerging SARS-CoV-2 variants necessitate understanding transmission in mixed-mode ventilated offices.
• Forecasting infection events (R-Event) is crucial for public health in shared indoor spaces.

Purpose of the Study:

• To develop and compare Artificial Neural Network (ANN) and Curve Fitting (CF) models for forecasting R-Event in mixed-mode ventilated offices.
• To investigate the relationship between CO2 levels and R-Event prediction.
• To assess model performance using various statistical indicators.

Main Methods:

• Real-time data collected in an Indian office environment during Spring/Summer 2022.
• Thirteen input features including environmental (temperature, humidity, CO2, AQI), occupancy (number, area/person, volume/person), and ventilation (opening area, fan speed, AC, outdoor wind/temp/humidity) parameters.
• ANN and CF models developed to predict R-Event; performance evaluated using correlation coefficient, RMSE, MAE, MAPE, NS index, and a20-index.

Main Results:

• The ANN model achieved a significantly higher correlation coefficient (0.9999) compared to the CF model (0.7439).
• ANN demonstrated superior accuracy and reliability in R-Event prediction over CF.
• CO2 levels were a key factor in forecasting infections within the office environment.

Conclusions:

• The proposed ANN model is a reliable and highly accurate tool for forecasting SARS-CoV-2 transmission events (R-Event) in mixed-mode ventilated office settings.
• ANN models offer a significant advantage over traditional CF methods for this predictive task.
• This research provides a valuable framework for infection risk assessment in indoor workspaces.