Akademik Veri Yönetim Sistemi
SOSYOEKONOMI, cilt.28, sa.46, ss.173-186, 2020 (ESCI)
Most of the studies on electricity consumption were conducted using econometric models and statistical methods. Studies aiming at predicting electricity consumption levels using machine learning methods based on household characteristics is a new approach in the literature. This study aims to present a model that predicts the electricity consumption levels in dwellings as lower consumption and higher consumption classes with the help of household and dwelling characteristics. Artificial Neural Networks were utilized as a machine learning method in the modelling phase. Data were gathered from Turkish Statistical Institution's Household Budget Surveys. The records having no electricity consumption were removed and the mean electricity consumption was determined from the remaining 32,765 households. Records above the mean were labelled as high-consumption class and that are below the mean were labelled as low-consumption class. ANN model training was carried out using 24,574 (70%) household data. Remaining 8,191 (30%) household data were used for testing the model. The success of the model was 75.11% at the training phase, 65.56% at the testing phase. As a result, the model proposal predicting electricity consumption levels using household and dwelling characteristics to contribute electricity production and distribution planning is presented.