Design, comparison and application of artificial intelligence predictive models based on experimental data for estimating carbon dioxide concentration inside a building

This paper describes applications of artificial intelligence to estimate indoor carbon dioxide levels, even in the absence of a direct CO2 sensor, based on simply obtained indoor parameters. The aim is to develop a novel universal model that is easy to use in a wide range of applications and easy to replicate. The presented predictive models are based on experimental data collected via affordable microclimate stations. They are built with Wi-Fi-enabled microcontrollers similar to Arduino, designed to monitor environmental factors such as temperature, humidity, human presence, atmospheric pressure, and carbon dioxide levels. The experimental data used to train and test the models are specifically collected for this analysis in a primary school classroom in Bialystok, Poland, through the homemade stations that gather data minute-by-minute. Linear and non-linear models include machine learning models such as SVM, random forest, decision tree, and neural networks. Among the various models tested, the random forest method, which relied solely on temperature, humidity, and human presence measurements, produced the most accurate results, achieving an R-squared value of 0.89. The use of temperature, humidity, and presence sensors, which are more affordable than CO2 sensors, highlights the novelty of the present analysis and the cost-effectiveness of predictive modelling in environmental monitoring.