Fraud detection in high voltage electricity consumers using data mining

This work presents a methodology and a computational system for fraud detection for high voltage electrical energy consumers using data mining. This methodology is based on a non-supervised artificial neural network called SOM (Self-Organizing Maps), which allows the identification of the consumption profile historically registered for a consumer, and its comparison with present behavior, and shows possible frauds. From the automatic consumer behavior pre-analysis, electrical energy companies will better direct its inspections, and will reach higher rates of correctness. The fraud detection system validation showed that the methodology is robust on the cases of lower consumption resulted by fraud, and on the cases of atypicality intrinsic to the consumer.