Predicting occupant behavior in office buildings based on thermal comfort variables using machine learning

Abstract

Office workers spend most of their time inside a building, and as a result, physical-environmental variables begin to play a crucial role in their productivity and performance. This study establishes a connection between machine learning models and the behavior of occupants and the self-assessed productivity they exhibit, through the use of various models. These models were implemented to identify and compare which of them better estimate this behavior, particularly the self-assessed productivity that individuals experience in their workplace. To accomplish this, physical-environmental variables, and the perceptions of occupants from various office buildings in the city of Concepción were collected. This study successfully compares the performance of four machine learning models (decision tree, K-Nearest Neighbor, Bayesian model, and neural network). Their performance was measured using indicators known as Accuracy, Precision, and Recall. These models were applied to both an original database and a balanced database, followed by a comparison of the results obtained. It can be established that there is a relationship between physical-environmental variables and the self-assessed productivity of workers. Furthermore, it can be mentioned that the neural network is the model that best describes this relationship and, therefore, achieves the highest performance. This study provides an approach to understanding occupant behavior from a machine learning perspective