A scalable Bluetooth Low Energy approach to identify occupancy patterns and profiles in office spaces

Building occupants are often assumed to follow deterministic schedules in building performance simulation programs. Therefore, to accurately capture the dynamic nature of the occupants' movement patterns, researchers have proposed various indoor localisation technologies to infer occupancy information with varying degrees of accuracy and resolution. Among these technologies, the Bluetooth Low Energy (BLE) technology emerged as a popular alternative due to its availability in smartphone devices, as well as its low cost and power demand. In this study, we proposed a scalable and less intrusive occupancy detection method that leverages existing BLE technologies found in smartphone devices to perform zone-level occupancy localisation, without the need for a mobile application. The proposed method uses a network of BLE beacons for data collection before passing the pre-processed data into a machine learning model to infer the occupants' zone-level location. A supervised ensemble model and a semi-supervised clustering model were proposed and evaluated to identify the best performing model. The feasibility of the proposed method is demonstrated during a five-week case study involving two office spaces in an academic building in Singapore. While the supervised ensemble model produced the best performance in terms of accuracy and macro-average f1-score, the semi-supervised model was able to produce a reasonable performance while using a fraction of the training data (