SUPERVISED MACHINE LEARNING FOR ELECTRICITY THEFT DETECTION: A SYSTEMATIC REVIEW OF TRENDS, CHALLENGES, AND FUTURE DIRECTIONS

Abstract

Electricity theft causes substantial financial losses and grid instability, requiring data-driven detection solutions, especially through supervised machine learning (SML) techniques. This review systematically examines 50 studies (2012–2024) from Scopus, IEEE Xplore, and Web of Science, following PRISMA guidelines to ensure a thorough and high-quality evaluation. It focuses on key themes such as algorithms, dataset challenges, performance metrics, and scalability, which are crucial for advancing electricity theft detection (ETD). SML methods such as decision trees, gradient boosting, and deep learning demonstrate high accuracy but encounter challenges, including inconsistent evaluation metrics, limited scalability for large-scale grids, and reliance on simulated datasets. Notably, only 28% of studies consider real-world scalability, and 54% lack robust evaluation metrics. Additionally, data issues such as missing values and class imbalance exacerbate these limitations. Future research should focus on developing hybrid models that effectively balance accuracy and interpretability, establishing standardized benchmarks for comparison, and incorporating real-world validation to enhance scalability and reliability in smart grid environments.