Electricity Theft Detection Using Machine Learning Approaches: A Case Study in Turkiye

Electricity theft, leading to financial losses and operational inefficiencies, is a significant challenge for energy utilities. In this study, advanced pre-processing, feature selection and model evaluation techniques were used to develop a machine learning model for detecting electricity theft. The dataset, which consists of over 53 million samples, was carefully preprocessed to eliminate missing values and irrelevant features. Principal component analysis (PCA) was applied to reduce dimensionality, and both undersampling and oversampling were used to avoid class imbalance. Four machine learning algorithms were evaluated: Random Forest, kNN, XGBoost and Logistic Regression. The training and evaluation of the models were performed in Jupyter Notebook using Joblib for efficient CPU-based parallel computation. The random forest with over-sampling achieved the highest performance with an accuracy of 98.23% and an F1 score of 0.90, showing the effectiveness of handling class imbalance. The results show that over-sampling the dataset leads to better results than under-sampling, emphasising the importance of this approach in detecting power theft.