Abstract
The illegal use of electricity, defective meters, and malfunctioning infrastructure is major cause of non-technical losses (NTLs) in electric distribution systems. Although machine learning techniques have been widely studied to solve this problem, a new framework is proposed to address the following challenges. (i) Given that fraudulent consumers remarkably outnumber nonfraudulent ones, the imbalanced nature of the dataset can have a major negative impact on the performance of supervised machine learning methods. (ii) Given a large number of dimensions present in the time series data used for training and testing classifiers, advanced signal decomposition techniques are required to extract the most relevant information. (iii) The effectiveness of classifiers must be evaluated using meaningful performance measures for imbalanced data. The core of our proposed framework contains two parts. First, we utilize Maximal Overlap Discrete Wavelet Packet Transform (MODWPT) for feature extraction from time-series data. Second, we use Random Under-Sampling Boosting (RUSBoost) algorithm for NTL detection. Moreover, our framework uses an extensive list of performance metrics to evaluate. Experiments demonstrate that the MODWPT combined with the RUSBoost algorithm can significantly improve the quality of NTL predictions.