Assessment of probabilistic models to estimate the occupancy state in office buildings using indoor parameters and user-related variables.

• An office was instrumented to collect human-activity and environmental parameters. • The ability of three models to estimate the occupancy state through the sensors data was verified. • Law of Total Probability, Naïve Bayes classifier, and CART were the chosen models. • Models were performed by considering 34 different combinations of parameters. • The number and the type of the used parameters influenced the models accuracies. Occupancy modeling in office buildings is still in progress and needs to be developed by using observable data. In this paper, an office building under the Mediterranean climate was instrumental in the collection of both indoor environmental parameters (air temperature and relative humidity, CO 2 , VOC) and user action-related variables (electricity power, window, door state, and air conditioning use). Each parameter was monitored along with the occupancy state at a one-minute time step for two years. The purpose of the investigation was to evaluate the potential application of three straightforward models, such as the Law of Total Probability (LTP), Naïve Bayes classifier (NB), and Classification and Regression Tree (CART), to estimate the occupancy state using the indoor measurements. Thirty-four (34) different combinations of parameters were applied on the developed models; the true positive rate (TPR), true negative rate (TNR), and accuracy were used as evaluation metrics. The results confirmed that the performances of the models were influenced by both the number and the typology of the used parameters. In particular, the CART model was found to be the least affected by them; almost half of the parameters' combinations provided accuracies higher than 93% and TNR higher than TPR. Accuracies of the order of 90% were obtained with NB and LTP. [ABSTRACT FROM AUTHOR]