Machine Learning to Predict and Forecast CO2 in New Zealand Classrooms

Indoor Air Quality (IAQ) monitoring conducted in school classrooms across New Zealand has revealed a prevalent failure to comply with domestic and global standards, especially regarding elevated CO2 concentrations. By leveraging an Internet of Things methodology, in conjunction with Machine Learning Modelling, we have successfully devised a model capable of forecasting CO2 levels in classrooms, predicting their non-compliance with New Zealand guidelines. These models are based on intensive data collection in 8 classrooms in the Wellington area. This collection includes data on occupancy, window and door opening, as well as temperature, relative humidity, and CO2 over the school day. The machine learning model is designed to predict CO2 considering other collected data. Evaluation of the key predictors of the model has been researched to understand the impact of the data collected on the model's performance. After establishing proof of concept through our initial research, we are expanding our monitoring and modelling efforts to encompass a wider range of classroom settings in New Zealand. At a wider range, this model could be used to classify classrooms in terms of their indoor environment, permitting a prioritisation for retrofitting the whole New Zealand classroom stock.