Occupancy detection from temperature, humidity, light, CO2 and humidity ratio measurements using machine learning techniques

Abstract

Order to save energy and to use energy resources efficiently, automatic occupancy determination based on sensor information is performed and energy is adjusted according to the demand in a closed area. In this study is used records consisting of T (temperature), H (humidity), L (light level), CO2 (carbon dioxide) and R (humidity ratio) sensor data. Occupancy analysis based on sensor data has been performed with REPTree (Reduced Error Pruning tree), NB (Naive Bayes), SVM (Support Vector Machine) and KNN (K Nearest Neighbor) classification algorithms. The highest classification success (97.98%) was obtained with the REPTree classification algorithm. Then the importance of the attributes is determined by the CFS (Correlation-based Feature Selection) algorithm to be taken into account in reducing costs in the data collection step and the effect of the attributes on the classification performance is examined. Finally, ensemble algorithms Adaboost, Bagging, RandomSubSpace are used to increase classification success and achieve more stable results. The performance evaluation criteria are shown that the ensemble algorithms improve the classification success. The highest classification success (98.28%) was obtained by using the Adaboost community algorithm together with the ADTree classifier. According to the sensor values in the dataset, the use of office room was determined with high success rate. © 2018 IEEE.