Electric Utility Customers’ Power Consumption Prediction Using Data Mining Techniques (A Case of Arba Minch District)

Abstract

Electric industry is one of most important service provider and back bone of the energy sector in the world. Ethiopia electric utility is the only national organization distributing electric power in our country. Electric power industries are being pushed to and quickly respond to the individual and organization needs and wants of their customers due to the dynamic and highly competitive nature of the industry. As the researcher noted in the literature, most previous studies developed Regression and classification models using few data set, two or less data mining algorithms, dataset also have less quality including inaccurate, inconsistent, incomplete and duplicated information. Therefore, the prime purpose of the current study was to develop a predictive model for anticipating the electric power consumption of EEU customers in Arba Minch district. The study employs and follow data mining techniques and methodologies to attain its objectives. Major steps followed in the research include: problem understanding, data understanding, data preparation, modeling, experimentation, and evaluation. The researcher used purposive sampling method. As a result, due to data availability, researcher exposure, and time constraint the Arba Minch District electric utility customers’ data covering six months (Sept. 2022 to Feb. 2023) was considered. The study used four predictive modeling algorithms based on their suitability to build both regression and classification tasks. Accordingly, the study included: Support Vector Machine, Decision Tree, Random Forest, and Artificial Neural Network algorithms for classification. The result achieved from the experiments showed that the Support Vector Machine algorithm performed best for regression task with root mean squared error (RMSE) of 350.12. Whereas, Decision Tree algorithm performed best on classification task with accuracy of 62.78%. This study regresses the power consumption value and also classify power consumption category as either high or low power consumption considering the nature of the customers. Finally, the study explored various power related problems and solutions from the analysis of the data and results of the current study.