A novel framework to detect electricity theft in regions with traditional metering; applying a hybrid machine learning-based approach to low-sampling-rate data

• The challenges posed by low-sampling-rate data are addressed using advanced machine learning and data processing techniques. • The issue of asynchronous consumption data readings is tackled using an integrative solution that includes pre-processing, over-sampling, and classification with the XGBoost algorithm. • Using a real dataset to evaluate the performance of the proposed approach, demonstrating effectively detection of electricity theft with high accuracy. Electricity theft poses a significant challenge to power distribution companies, leading to substantial economic losses and threatening the stability and reliability of power systems worldwide. This study addresses the complexities of detecting electricity theft in regions that rely on traditional meters, which record low-sampling-rate data. Due to the asynchronous nature and low frequency of data collection, existing machine learning models often struggle with noise and class imbalance, resulting in reduced detection accuracy. To overcome these challenges, a novel framework developed in this article that combines advanced preprocessing techniques with machine learning. The developed framework includes data conversion to monthly averages, normalization, and the application of the Synthetic Minority Over-sampling Technique to balance the dataset. An Extreme Gradient Boosting algorithm is then applied for effective electricity theft detection. The study utilizes a real dataset from a northern Iranian province, comprising 119,131 customers with meter readings taken at 40 to 60 days intervals. The proposed approach offers a promising solution for electricity theft detection in regions employing traditional metering systems, with the accuracy and precision of 99.7 % and 98.2 %, respectively.