Futureproofing and scaling machine learning for occupancy prediction

Abstract

An important instrument for achieving smart and high-performance buildings is Machine Learning (ML). A lot of research was done in exploring the ML learning models for various applications in the built environment such as occupancy prediction. Nevertheless, this research focused mostly on analyzing the feasibility and performance of different supervised ML models but have rarely focused on practical applications and scalability of those models. In this study, we are proposing a transfer learning method as a solution to few typical problems with the practical application of ML in buildings. Such problems are scaling a model to another (different) building, collecting ground truth data necessary for training the supervised model and adapting the model when conditions change. The practical application examined in this work is a deep learning model used for predicting room occupancy using indoor air quality (IAQ) IoT sensors. The importance of occupancy prediction has risen in recent times of remote work and is especially important for futureproofing of the built environment. This work proves that it is possible to reduce significantly the need for ground truth data collection for deep learning based occupancy detection model. Additionally, the robustness of the transferred model was tested, where performance stayed on similar level if suitable normalization technique was used.