Abnormal consumption analysis for fraud detection: UTE-UDELAR joint efforts

Within the framework of the Energy Recovery Unit of the Technical-Commercial Service of Montevideo, UTE, for the reduction of Non Technical Losses, a research project was carried out jointly with the Institute of Electrical Engineering of UDELAR. The project had the aim of designing different strategies of automatic classification that separate normal consumption measurements from abnormal ones which represent clues of possible sources of Non Technical Losses. Different classifiers were implemented and several field tests were conducted, with promising results. Several criteria for the incorporation of new features are proposed in this work. These criteria are complementary to those derived from consumptions. An analysis of the performance of said features was conducted, showing that improving classifier performance is possible with this method.