Tree based classification methods for occupancy detection

Latest smart buildings are not only be intelligent to allow occupant to control the light, heating, cooling, gas and other systems but also focuses on occupancy detection since accurate occupancy detection can result in saving energy up to 42% as can be seen in literature. For this aim, different autonomous systems including sensors, actuators, microcontroller and etc. are at the development phase for smart buildings. At this point, determination of classification methods to detect the occupancy together with hardware plays crucial role. Having done in this study 3 different classification methods that is based on Machine Learning Methods were applied on benchmark dataset named Occupancy by UCI Machine Learning Repository, 2016. The classifiers are Random Forest, Decision Tree and Bagging. They were chosen by following two principals. First one is to have classifier methods that were not use in literature for benchmark dataset and the second one almost never usage of tree based classifiers in the literature. The Occupancy dataset consists of light, temperature, humidity, CO2 and occupancy. It has been seen that the highest accuracy or prediction ratio was obtained as 99, 368% by Decision Tree method namely; Random Forest. This result was compared with the results of the studies on the same benchmark dataset. It has been seen that it is the second best accuracy ratio after Fuzzy Granulation (Fgf) method among 16 different Machine Learning Based classification methods. Additionally, Decision Tree and Bagging had the accuracy ratio of 99.222% and 99.207, respectively. These ratios are also higher than other methods used in literature but Fgf. Thus this study showed how decision tree can be promising for occupancy detection.