Rapid prediction of indoor air quality index using artificial neural network (ANN) in educational spaces: a case study of university in Thailand

Indoor air quality significantly affects educational environments where students spend prolonged hours in enclosed classrooms. Poor indoor air quality causes sick building syndrome, respiratory issues, and reduced learning performance. Despite extensive outdoor air quality research, limited studies focus on predicting indoor air quality index (IAQI) using outdoor parameters. This study developed and validated a deep artificial neural network to predict IAQI in educational buildings using outdoor air pollutants. A comprehensive study was conducted from June to September 2024 in four rooms in educational building in Pathumthani, Thailand. Indoor and outdoor air parameters were measured using calibrated instruments: Aeroqual AQM 64 , Particle Count PC-4000, GASMET GT5000, and Q-TRAK Model 7575. Specific pollutants measured included particulate matter (PM1, PM2.5, PM10), carbon dioxide, carbon monoxide, nitrogen oxides, ozone, total volatile organic compounds, temperature, relative humidity, and air change rate. Data collection involved 5-minute interval measurements during working hours, yielding 240 data points per room (total n=960). A multilayer perceptron deep artificial neural network was developed using WEKA software. Each room's dataset was divided into training (70%, n=168) and testing (30%, n=72) sets. Network architecture consisted of 3-4 input variables, 1-2 hidden layers with 1-10 nodes, and one output layer. Training continued for 100,000 epochs with learning rates of 0.05-0.5. IAQI was calculated using U.S. EPA methodology, with model performance evaluated using Mean Absolute Percentage Error. Pearson’s correlation analysis revealed significant associations between outdoor pollutants and IAQI, with room-specific variations in predictor variables. The deep neural network models demonstrated excellent predictive performance with MAPE values of 0.21-0.64%, well below the 10% threshold for high accuracy. This study successfully demonstrated that deep artificial neural networks can accurately predict indoor air quality using outdoor parameters, providing a cost-effective monitoring alternative for educational institutions. Therefore, the use of this prediction model is recommended for evaluating the indoor air quality index in educational settings, owing to its simplicity, accuracy, and time-efficient nature