People tend to spend the majority of their time indoors. Indoor air properties can significantly affect humans' comfort, health, and productivity. This study utilizes measurement data of indoor conditions in a kindergarten in Sofia, Bulgaria. Autoregressive integrated moving average (ARIMA) and long short-term memory (LSTM) recurrent neural network (RNN) models were developed to predict CO2 levels in the educational facility over the next hour based on 2.5 h of past data and allow for near real-time decision-making. The better-performing model, LSTM, is also used for temperature and relative humidity forecasting. Global comfort is then estimated based on threshold values for temperature, humidity, and CO2. The predicted R2 values ranged between 0.938 and 0.981 for the three parameters, while the prediction of global comfort conditions achieved a 91/100 accuracy.
Keywords: air quality; comfort conditions; indoor environment; machine learning; predictive modeling.