Extraction des connaissances à partir d’une base de données (application à la détection de fraude dans la consommation d’électricité et du gaz)

Abstract

Our work involves combining the fields of knowledge extraction and artificial intelligence to develop a sophisticated system for detecting fraud related to electricity and gas consumption. To achieve this, we leveraged the online database of STEG and implemented two state-of-the-art classifiers, XGBoost and LightGBM, which are among the top machine learning algorithms for solving such problems. After conducting a comparative evaluation, we selected the most performant model, LightGBM, taking into account several metrics demonstrating its superiority. Using this model, we achieved a precision rate of 95.00% and a score of 88.71%, resulting in an excellent 19th position out of 295 participants in the Zindi challenge. These results showcase the effectiveness of our innovative approach and our dedication to tackling challenges in energy fraud detection.