Electricity Theft Detection using Pipeline in Machine Learning

Electricity theft is the primary cause of electrical power loss that significantly affects the revenue loss and the quality of electrical power. Nevertheless, the existing methods for the detection of this criminal behavior of theft are diversified and complicated since the imbalanced nature of the dataset, and high dimensionality of time-series data make it challenging to extract meaningful information. This paper addresses these problems by developing a novel electricity theft detection model, integrating three algorithms in a pipeline. The proposed method first applies the synthetic minority oversampling technique (SMOTE) for balancing the dataset, secondly integration of kernel function and principal component analysis (KPCA) for the feature extraction from high dimensional time-series data, and support vector machine (SVM) for the classification. Besides, the performance of the proposed pipeline is measured using a comprehensive list of performance metrics. Extensive experiments are performed by using real electricity consumption data, and results show that the proposed method outperforms other methods in terms of theft detection.