Detecção de anomalias em telecomunicações através de um sistema baseado em conhecimento que utiliza consulta por similaridade, DWT e RDR como ferramentas de apoio

Abstract

Telephony fraud is a world phenomenon. Current estimates show losses of from 15 to 55 billion dollars a year in the telephony industry, which runs businesses around 1.5 trillion dollars [42]. Fraud detection is a critical point for these companies. Fraud risks force companies to make a great effort to analyze clients' calls traffic. This paper proposes a Knowledge System (KS) that performs similarity search in time series with the purpose of detecting anomalies, in advance, in the clients' call traffic. Since there is a great amount of data (call) to be analyzed, the system uses Haar Wavelets as a dimensionality reduction technique to improve the eficiency of the searches. For the purpose of indicating the actions that should be taken as a result of similarity searches, a knowledge system based on production systems and on case based reasoning system is utilized. This hybrid system is called Ripple Down Rules (RDR).