•  
  •  
 

Abstract

Electricity fraud detection presents a significant challenge for power distribution companies, as non-technical losses resulting from fraudulent activities adversely affect revenue and operational efficiency. This study presents a machine learning-driven approach for accurately identifying fraudulent electricity consumption patterns. A comprehensive analysis of electricity usage data is conducted using multiple classification models, including Support Vector Machine (SVM), K-Nearest Neighbors (KNN), a Stacking Model integrating SVM, KNN, Random Forest, and Gradient Boosting, as well as a Weighted Model leveraging confidence-based prediction adjustments. Model performance is evaluated using key metrics, including accuracy, precision, recall, F1-score, AUC-ROC, and confusion matrices. The results indicate that the Stacking Model achieves the highest predictive performance, with an AUC-ROC of 1.000, while the Weighted Model effectively balances precision and recall, attaining an F1-score of 0.9909. Furthermore, ROC curves and confusion matrices illustrate each model's effectiveness in differentiating fraudulent from legitimate customers. The findings of this study culminate in the development of an application designed for real-world deployment, enabling electricity providers to detect and mitigate fraudulent activities.

Article Type

Article

Download

Included in

Engineering Commons

Share

COinS