Near Zero-Energy Home Prediction of Appliances Energy Consumption Using the Reduced Set of Features and Random Decision Tree Algorithms

Abstract

This paper presents methods for prediction of energy usage of different appliances in homes. Dataset comprising 14804 samples include measurements of weather from a nearby airport station, temperature and humidity sensors from a wireless network and recorded energy use of lighting fixtures. These measurements are sorted into 32 features, from which 17 were filtered and showed to be sufficient for energy usage prediction. Two methods for prediction were trained and tested: Random forest and Random tree. The performance of the methods was studied and it has been showed that the random forest gives better results than random tree method and that it has good performance in prediction of energy use of appliances.