Privacy-Preserving Sensing and Two-Stage Building Occupancy Prediction Using Random Forest Learning

Sensing and predicting occupancy in buildings is an important task that can lead to significant improvements both energy efficiency occupant comfort. Rich data streams are now available allow for machine learning-based algorithm implementation of direct indirect estimation. We evaluate ensemble models, namely, random forests, on collected from 8×8 PIR matrix thermopile sensor with the dual goal individual cell temperature values subsequently detecting status. Evaluation method based a real case study deployed IT Hub Bucharest, which we have over three weeks ground data, analyzed, used it order predict room. Results show 2–4% mean absolute percentage error prediction id="M2">> 99% accuracy three-class model detect human presence. The resulting outputs be by predictive building control models optimize commands various subsystems. By separating specific deployment system architecture structure, application easily translated other usage profiles built environment entities. As compared vision-based systems, our solution preserves privacy improved performance when single or