Research Article
Privacy-Preserving Sensing and Two-Stage Building Occupancy Prediction Using Random Forest Learning
Table 1
Taxonomy of sensing platforms and occupancy methods for space management.
| Sensor type | Occupancy method | Experiment duration | Location type | Algorithm | Source |
| PIR | Presence prediction | 50 hours | Several offices | Infinite hidden Markov model | [7] | CO2 sensors + others: light, PIR, acoustic | Occupancy detection | 7 days | 1 cubicle | Decision trees | [8] | Wi-Fi | Occupancy counting | 1 week | 2 lecture rooms | Newton Interp. + NN model | [9] | Distributed plug load power strip sensors | Occupancy detection | 2 weeks | 3 rooms | Bayesian inference, graphical lasso, influence model | [10] | PC23D stereo cameras | Occupancy counting | 15 days | 4 rooms | PLCount | [11] | PIR + infrared sensor | Occupancy counting | 3 weeks | 10 building areas | KNN | [12] | Temperature, humidity, light, CO2 and digital camera temperature, motion sensor, RFID tags | Occupancy detection | 1 month | 1 office | Random Forest, GBM, LDA, CART | [13] | Occupancy prediction | 61 days | 5 homes | Mean of nearest past days | [14] | CO2 sensors | Occupancy counting | 4 months | 2 rooms office and theatre | Seasonal trend decomposition | [15] | Wi-Fi | Occupancy detection | n/a | 1 conference room | Random Forests | [16] | PIR matrix | Occupancy detection activity recognition | n/a | Laboratory | Fuzzy background removal | [17] | PIR, CO2, power water, noise | Occupancy prediction | n/a | Office apartment multizone house | Bayesian networks | [18] |
|
|