Prediction of electricity consumption based on DT and RF: An application on USA country power consumption

Non-technical losses, especially by cause of electricity theft have been a primary care in power system during long period of time. Huge absorption of electricity in a fraudulent way may mismatch the gap between supply and demandprofoundly. Detection of electricity theft is done in two levels, data processing and analysis scheme. This paper presents the first level of data processing to predict the expected electricity consumption of home by using suitable supervised machine learning algorithms particularly random forest and decision tree. Random forest and decision tree are extremely attractive machine learning algorithms for solving binary and continuous type problems. These works equate the classification performance of DT and RF which was generated using 100 trees. The estimation of electricity consumption of data consists of 237 data samples for the prediction of electricity was accomplished by using the MATLAB. The data came from energy department in United States of America for the year 2014. For assessing the performance of decision tree and random forest misclassification error rate, receiver operating characteristics and prediction accuracy was used as performance measure. An experimental completion expose that the random forest algorithm gives better prediction accuracy (95.78%) in comparison with DT (91.6%) and the error rate of RF and DT are 0.197,0.906 respectively.