The aim of this study is to predict the CO2 concentration levels based on data obtained in the auditorium of the new university building of the Faculty of Architecture, Civil Engineering and Geodesy and Faculty of Forestry (FACEG-FF) at the University of Banja Luka in correlation to the operation of the mechanical ventilation system. The data used for the analysis were collected from measurements and surveys on students’ subjective perception of indoor thermal and air quality. Data analysis and statistical processing were carried out using Statistical Product and Service Solutions (SPSS). The influence of measured physical parameters and the impact of user behaviour on CO2 concentration using the chi-square automatic interaction detection (CHAID) decision tree technique in a university auditorium was predicted. The dataset comprised 11 input variables (A1-A11), a continuous output variable representing CO₂ concentration (B1, ppm), and two predicted categorical outputs: B2, describing CO₂ concentration classes, and B3, indicating air quality satisfaction based on a 1000 ppm CO₂ threshold. Classification accuracy was between 89.2% and 97.5% for dependent variable B2 (class of indoor air quality) and 94.2% and 99.2% for dependent variable B3 (satisfactory - value below or above 1000 ppm CO₂). The findings indicate that maintaining CO₂ levels below 1000 ppm is crucial for ensuring satisfactory indoor air quality and thermal comfort among occupants.