Climate change is causing an increasing cooling demand in residential buildings. Understanding the drivers behind occupants’ use of air conditioning is critical for accurate building energy performance analysis. However, occupant-building interactions are highly variable and influenced by multidisciplinary factors, which cause critical uncertainty in behavioural modelling and energy use prediction. This study proposes the use of GLMMs to investigate if the inclusion of multi-domain factors (including physical, behavioural, and contextual domains) in behavioural models increase the predictive performance, in comparison with single-domain models. Results from a monitoring campaign in three residential building apartments reveal a better performance of the multi-domain model in predicting occupant behaviour. Insights obtained from the multi-domain model then reveal that daily variability and apartment differences significantly influence air conditioning status. Occupancy, outdoor humidity, and CO2 levels increase the likelihood of activation, while high air temperature differences between indoors and outdoors, high indoor humidity and window opening reduce it.