Accurate Occupancy Detection via Label Noise Filtering Technique

An accurate occupancy detection is the inevitable part to save the energy consumption in an office room or a building. Recently, a new approach which uses statistical machine learning models has been proposed to detect the occupancy more accurately. This approach is proven to be a quite successful, however, there is still room to enhance the accuracy more by considering the point that the data gather by sensors always contains the noise. In this paper, we introduce the label noise filtering method which eliminates this suspicious noisy data and show great enhancement of detection accuracy just by eliminating the noisy. The evaluation results indicate that the accuracies of occupancy detection have increased by 1.5% in average (from 97.4% to 98.9%) just by filtering the suspicious data. Especially in the classification and regression tree (CART) model, the noise filtering greatly increases the accuracy from 94.3% to 97.6%.