Smart Meter Electricity Theft Detection Using Machine and Deep Learning Approaches

Batir, Serap (2025) Smart Meter Electricity Theft Detection Using Machine and Deep Learning Approaches. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (1MB) | Preview
Preview	PDF (Configuration Manual) Download (338kB) | Preview

Abstract

Electricity theft is a growing problem for utility companies all over the world. This problem has been shown to cause significant loss of money and inefficiency in operations. This study analysed the use of machine learning and deep learning methods in identifying fraudulent consumption behaviour in smart meter data. The study employed a scenario-based approach, which creates flat reduction, selective tampering and consumption mean manipulation scenarios and used them in the training and evaluation of detection problems. Comparative experimentation showed that Long Short-Term Memory and Gated Recurred Unit (GRU) models perform better (94% accuracy) in the identification of temporal patterns, whereas Random Forest classifiers were highly effective (94% accuracy). The research reveals major drawbacks in the conventional methods, as basic K-nearest Neighbors were less efficient in dealing with dependencies over long periods and Support Vector Machines had poor scaling. XGBoost incurred the highest training time with only moderate accuracy gains. To summarize, this paper brings value to the field because it proposes an evaluation framework to be applied to theft detection algorithms, which has been tested with realistic consumption data. The study also sets down workable criteria on model selection based on the needs of an operation, given the balance between accuracy and resources. This contribution shows which methods work well in detecting electricity theft and highlights some of main challenges in implementing them.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Rustam, Furqan UNSPECIFIED
Uncontrolled Keywords:	Electricity theft; deep learning; smart meter; machine learning; anomaly detection
Subjects:	T Technology > TK Electrical engineering. Electronics. Nuclear engineering > Electricity Supply Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning
Divisions:	School of Computing > Master of Science in Data Analytics
Depositing User:	Ciara O'Brien
Date Deposited:	24 Aug 2026 15:49
Last Modified:	24 Aug 2026 15:49
URI:	https://norma.ncirl.ie/id/eprint/9619

Actions (login required)

View Item