Electricity Theft Detection using Binary-Class Imbalanced Classification

Electricity theft refers to the unauthorised consumption of electricity. The energy sector faces a notable challenge of non-technical losses, which mainly arise from theft, and these losses have severe consequences for power suppliers. These effects include decreased revenue for suppliers and increased expenses for utility services for consumers. The energy supply’s stability has been compromised due to unauthorised energy consumption despite the simultaneous increase in energy production. The present circumstances significantly affect the reliability of the electrical infrastructure, leading to increased expenses for law-abiding users. Preventing unlawful energy use is the highest priority in nations like Bangladesh, particularly given the significant financial implications of energy production. This study presents a methodology that uses supervised machine learning techniques to identify non-technical losses in imbalanced datasets effectively. Several techniques are employed to address the problem of imbalanced datasets, including oversampling methods such as random oversampling, synthetic minority oversampling technique (SMOTE), support vector machine SMOTE (SVM SMOTE), and borderline SMOTE. Undersampling methodologies, namely random undersampling, cluster centroid, edited nearest neighbour, and neighbourhood cleaning rule, are also used. The present study employs several kinds of machine learning techniques, including K-Nearest Neighbour, Logistic Regression, Support Vector Machine (SVM), and Decision Tree, to analyse consumption data obtained from a sample of 56,618 records spanning the years 2017, 2018, 2019, and 2020.