Forecasting Indoor Air Quality Using Machine Learning Models

Abstract

As people typically spend a significant portion of their time indoors, indoor air pollution is the primary cause of nausea, dizziness, headaches and other health issues. Therefore, indoor air quality (IAQ) monitoring and prediction is important to protect people from indoor air pollution. The indoor pollutant prediction can be efficiently tackled by using machine learning (ML) models. This paper focuses on predicting IAQ based on several important pollutants including CO2, humidity, PM10, PM2.5, temperature, and volatile organic compounds (VOC). In particular, we evaluate and compare eight ML models namely Light Gradient Boosting Machines (LightGBM), eXtreme Gradient Boosting (XGBoost), Random Forest (RF), K-Nearest Neighbors (KNN), Support Vector Regression (SVR), Decision Tree (DT), Linear Regression (LR), and Long Short-Term Memory (LSTM). These ML models are trained and then predict pollutants on GAMS dataset which is not well-investigated in literature. The evaluation of the models employs standard measures such as mean square error (MSE) and mean absolute percentage error (MAPE). Our results highlight LightGBM, SVR, and XGBoost as optimal models for IAQ prediction. Specifically, LightGBM achieves an impressive CO2 prediction MAPE score of 0.0960%. SVR demonstrates strong MAPE scores for humidity (0.0185%), temperature (0.0264%), and VOC (3.1953%) predictions. XGBoost attains notable MAPE scores of 0.0414% and 0.0399% for PM10 and PM2.5 models respectively. Thus, advocating for their application across diverse settings—residences, offices, educational institutions, and healthcare facilities—to forecast and monitor IAQ, the study contributes to mitigating health risks tied to indoor air pollution.