Predicting Airborne Pollutant Concentrations and Events in a Commercial Building Using Low-Cost Pollutant Sensors and Machine Learning: A Case Study
Published in Building and Environment, 2022
Recommended citation: Mohammadshirazi, A., Kalkhorani, V. A., Humes, J., Speno, B., Rike, J., Ramnath, R., & Clark, J. D. (2022). Predicting airborne pollutant concentrations and events in a commercial building using low-cost pollutant sensors and machine learning: a case study. Building and Environment, 213, 108833. https://www.sciencedirect.com/science/article/pii/S0360132322000816
Summary
This paper explores the prediction of indoor airborne pollutant concentrations to enhance smart indoor air quality control, addressing four key objectives: 1) Assessing the efficacy of various low-cost sensors for air quality prediction, 2) Identifying the most effective predictive algorithms, 3) Exploring the potential for forecasting future indoor pollutant levels, and 4) Developing techniques for anticipating high concentration events. The study rigorously compares four prediction methods (Rolling Average, Random Forest, Gradient Boosting, Long-Short Term Memory) across eight different pollutants in a commercial building in California. It concludes that Long-Short Term Memory generally outperforms other methods, with the best sensor combinations varying according to the specific pollutant. Additionally, the research finds that using a regression-based approach to classify elevated concentration events is marginally more effective than direct classification methods for most pollutants.
Authors
Ahmad Mohammadshirazi, Vahid Ahmadi Kalkhorani, Joseph Humes, Benjamin Speno, Juliette Rike, Rajiv Ramnath, Jordan D. Clark