Online Learning Techniques for Occupancy Detection on Resource Constrained Devices

In light of the increasing impact of climate change, the demand for energy-efficient solutions across various sectors is becoming crucial. This paper explores the feasibility of utilizing Online Learning techniques to detect room occupancy in smart buildings, addressing the challenges of privacy and cost. Our methodology leverages existing environmental sensor data without requiring additional hardware, ensuring privacy and cost efficiency. The adaptability of Online Learning models to new environments is a key advantage, eliminating the need for extensive data storage and making the approach suitable for real-world applications. This research focuses on developing and testing this innovative approach, emphasising the constraints of embedded devices to ensure practical deployment without new hardware requirements.
Our extensive testing campaign evaluates the performance of the five most commonly used algorithms in this field, highlighting the viability of Online Learning techniques for deployment on embedded devices. This research lays the foundation for developing energy-efficient solutions in smart buildings, offering a cost-effective and privacy-oriented approach to occupancy detection.