Application of supervised machine learning approaches for predicting household electricity consumption in Alexandria, Egypt

Abstract

The electricity consumption in the residential sector, which currently represents about 27% of the world’s electricity consumption, has been experiencing continued growth due to economic and population growth. Therefore, proper management of future electricity provision entails predicting consumption patterns in the future to address increasing demand. For this purpose, Machine learning algorithms can support proper management of supply and demand for electricity through providing more accurate predictions of electricity consumption. However, the scarcity of data on electricity consumption patterns and their determinants, is one of the challenges that may restrict the potential of ML in predicting electricity consumption. This paper is intended to develop a machine learning based approach for predicting electricity consumption in the residential sector at the household level in Alexandria, Egypt, under data scarcity. For this purpose, the oversampling technique is applied to overcome data scarcity. It is found that bagging classifier, decision tree classifier, random forest classifier, and gradient boosting classifier have the highest performance with average accuracy exceeding 80%. This indicates that supervised machine learning algorithms that are tree-based structure gave higher accuracies for predicting seasonal household electricity consumption in the residential sector. Random forest is selected to develop an ML model for predicting electricity consumption at household level.