Electricity theft is a widespread problem with significant negative economic and financial impacts. This problem is still a challenge for energy utilities all around the world. Estimates place the cost of power theft and fraud in the energy sector at $96 billion yearly, with costs of up to $6 billion in only the United States, making it the third largest form of theft in the country. Therefore, tackling electrical fraud and theft is more necessary now than ever.
Different methods are being used or developed to detect this practice and lessen its effects. The project aims to detect electricity theft and fraud through machine learning by analyzing customers’ consumption patterns, among other features like bill history, reading remarks, and regions. The dataset is provided in Kaggle by the Tunisian Company of Electricity and Gas (STEG), containing 43 years of records of more than 135,000 customers with 21 different attributes. The project will adopt CRISP-DM, as the methodology used for the project completion, which provides a structured approach to data mining project planning,
The adapted supervised machine learning models for this project were decision tree, random forest, and support vector machine since they are considered the most common models used in fraud detection based on the conducted literature review. The final model selection was based on different metrics, the accuracy, precision, recall, and F1 score of the model. The random forest model surpassed the other two models, achieving an accuracy of 82.73%.
Fraud detection will reduce the significant impact of financial losses and enhance the service quality provided by electric utilities. Furthermore, the obtained results from the selected model can be used by electric utilities, especially in developing countries, to prioritize their Advanced Metering Infrastructure (AMI) installation plans in areas where fraud is detected, which will smoothen the transformation journey financially.
Professional Studies (MS)
Department, Program, or Center
Graduate Programs & Research (Dubai)
Rajab, Ali Jassim, "Electricity Theft and Energy Fraud Detection" (2023). Thesis. Rochester Institute of Technology. Accessed from