Look twice: Uncover hidden information in room climate sensor data

Connected sensors are on the march to become pervasive. While they are often deployed for a single purpose it is worth to take a second look. In this study, we show that the widespread Netatmo weather station which is intended to monitor and improve indoor climate can be used to estimate binary occupancy of individual rooms. We collected data from 11 rooms in 3 apartments including binary occupancy for several days. We show that CO₂ measurements and derivatives thereof qualify as observables to be used in Hidden Markov Models and achieve accuracies well above 75% in most cases. However, we see that the accuracy metric is often misleading for such timeseries data and consider additional performance metrics as well which show varying results depending on the respective occupancy patterns of a room.