Accurate occupancy detection of an office room from light, temperature, humidity and CO2 measurements using statistical learning models

The accuracy of the prediction of occupancy in an office room using,data from light, temperature, humidity and CO2 sensors has been evaluated with different statistical classification models using the open source program R. Three data sets were used in this work, one for training, and two for testing the models considering the office door opened and closed during occupancy. Typically the best accuracies (ranging from 95% to 99%) are obtained from training Linear Discriminant Analysis (LDA), Classification and Regression Trees (CART) and Random Forest (RF) models. The results show that a proper selection of features together with an appropriate classification model can have an important impact on the accuracy prediction. Information from the time stamp has been included in the models, and usually it increases the accuracy of the detection. Interestingly, using only one predictor (temperature) the LDA model was able to estimate the occupancy with accuracies of 85% and 83% in the two testing sets. (C) 2015 Elsevier B.V. All rights reserved.