Electricity Consumer Category Classification With Supervised Machine Learning Techniques Using Consumption Information

Electricity distribution utilities aim to reduce technical and non-technical losses (TL & NTL) while maximizing revenue through accurate tariff assignment. For consumers using smart meters, manual meter reading is unnecessary. Although, for those using conventional digital or analog meters, meters are checked by utility staffs regularly, still there persists some wrongly assigned tariffs. Identifying those misassigned tariffs is extremely challenging without manual intervention. This study focuses on classifying residential and commercial consumers using their seasonal usage patterns, sanctioned phase, and load. We compared various machine learning models—Logistic Regression, K-Nearest Neighbors (KNN), Decision Tree, Random Forest, and AdaBoost—and introduced a hybrid model combining Random Forest and AdaBoost with KNN as the meta learner. Our results show that the proposed hybrid model achieves the highest performance, with 86% accuracy, highlighting its potential for improving tariff classification based on meter consumption data.