Smart Electricity Theft Detection Using Behavioural Feature Engineering and Machine Learning

Electricity theft is a common problem for electricity distribution companies and can result in financial losses and difficulties in managing electricity supply. Traditional methods of detecting electricity theft mainly depend on manual inspection, meter checking and predefined rules. These methods may take more time and may not identify unusual electricity consumption behaviour effectively. In this study, Machine Learning is used to identify possible electricity theft based on consumer electricity consumption patterns. The study uses the SGCC Electricity Theft Detection dataset containing 42,372 consumer records with electricity consumption information over different dates. Behavioural features such as average consumption, consumption variation, zero-consumption days, nonzero consumption and missing-consumption patterns are created from the original data. Five Machine Learning models, namely Decision Tree, Random Forest, Support Vector Machine, XG Boost and Multi-Layer Perceptron, are developed and compared..