Predicting hourly indoor ozone concentrations with sensor-based measurements and easily accessible predictors

Area of Science:

• Environmental Science
• Public Health
• Data Science

Background:

• Indoor air quality, particularly ozone (O3) levels, significantly impacts human health.
• Few studies have utilized machine learning for predicting indoor O3 concentrations.
• Accurate indoor O3 exposure data is crucial for epidemiological studies.

Purpose of the Study:

• To develop and validate machine learning models for predicting hourly indoor O3 concentrations.
• To assess the impact of easily accessible predictors, including ventilation behavior, on indoor O3 prediction accuracy.
• To establish a scalable method for indoor O3 exposure assessment across diverse geographic locations.

Main Methods:

• Collected hourly indoor O3 data using low-cost sensors across 18 cities in China.
• Utilized ambient O3 concentrations, meteorological factors, and window status (ventilation proxy) as predictors.
• Developed and evaluated random forest models to predict indoor O3 levels.

Main Results:

• The inclusion of window status as a predictor enhanced model performance, increasing R2 from 0.80 to 0.83 and reducing RMSE from 7.89 to 7.21 ppb.
• The models accurately captured hourly variations in indoor O3 concentrations.
• Indoor O3 levels were consistently lower and more stable than outdoor concentrations.

Conclusions:

• Ventilation behavior, indicated by window status, is a critical factor in improving the accuracy of indoor O3 prediction models.
• Machine learning models using accessible variables offer a promising approach for large-scale indoor O3 exposure assessment.
• Sole reliance on ambient O3 data may lead to inaccurate personal exposure estimations, highlighting the importance of indoor monitoring and prediction.