Main Article Content
At present, there is a high incidence of telecommunications fraud in the world, and the overall form of anti telecommunications fraud is still severe. With the continuous evolution of cutting-edge technologies such as big data and artificial intelligence, new solutions are given to tap the characteristics of telecommunications fraud and improve the accuracy and coverage of anti fraud model identification. This paper uses the desensitized signaling data, voice message details, APP traffic data and billing accounting data to analyze the characteristics and preferences of telecommunications fraud users. Through the experimental comparison of machine learning(ML) and deep learning(DL) model classification, this paper explores the factors to improve the accuracy of model classification, and finally verifies and expounds the feasibility of prediction model selection through model training and testing.