A comparative study on optimized machine learning and deep learning models for the detection of electricity theft

Saha, Oindrila (2023) A comparative study on optimized machine learning and deep learning models for the detection of electricity theft. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (1MB) | Preview
Preview	PDF (Configuration manual) Download (2MB) | Preview

Abstract

The act of stealing electricity is a major obstacle for utility providers on a global scale, resulting in enormous financial losses and jeopardizing the integrity of the power infrastructure. In this research, a variety of machine learning and deep learning models, combined with Particle Swarm Optimization (PSO), are utilized to conduct a comparative analysis of electricity theft detection. The main thrust of this paper is the application of PSO to improve the parameters of every model, thereby improving their predictive capabilities. This ground-breaking combination of PSO with multiple models provides significant enhancements in both precision and productivity, constituting an innovative contribution to the field. The proposed methodology includes the implementation of XGBoost with PSO, Random Forest with PSO, Decision Tree with PSO, CNN with PSO, and LSTM with PSO. This report evaluates the accuracy of each model to determine the optimal one through extensive training and testing. The classification reports offer crucial performance indicators, with a focus on accuracy, recall, precision, and F1-score. Notably, the XGBoost with PSO, Random Forest with PSO, and LSTM with PSO models stand out as the top performers, reaching an astounding accuracy of over 80%. These advanced models have exceptional ability in managing intricate, unbalanced datasets that are crucial in fraud identification. To summarize, the hybrid ML and DL techniques demonstrates great potential in improving the detection of electricity fraud. Potential future pursuits may involve delving into more optimization approaches, incorporating varied deep learning architectures, implementing real-time systems, and expanding the incorporation of broader datasets. This study establishes a foundation for novel approaches to protect power distribution systems against fraudulent actions and new threats. Utility firms can effectively utilize these models by analyzing real electricity usage data to detect instances of electricity theft and mitigate significant revenue losses in the power sector.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Hafeez, Taimur UNSPECIFIED
Uncontrolled Keywords:	Electricity Theft Detection; Hybrid Models; Particle Swarm Optimization; XGBoost with PSO; Random Forest with PSO; Decision Tree Classifier with PSO; CNN with PSO; LSTM with PSO
Subjects:	H Social Sciences > HV Social pathology. Social and public welfare > Criminology > Crimes and Offences Q Science > QA Mathematics > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science 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:	Tamara Malone
Date Deposited:	13 Jan 2025 10:03
Last Modified:	13 Jan 2025 10:03
URI:	https://norma.ncirl.ie/id/eprint/7312

Actions (login required)

View Item