Fraud risk management in mobile cellular communications systems

Performance of the risk management system that is used for the detection of fraud on GSM Systems is investigated. In order to improve the current system performance, 4 different data mining methods is used. These methods are K-Nearest Neighbour, Naive Bayes, Random Forest, Support Vector Machine.Data Set is, 5641 GSM number belonging the X GSM operator. Suspected of fraud, X operator deactivated (service deactivated or full deactivated) these GSM numbers during 2 months. For 5641 GSM numbers, 76 feature is created and classifed with 4 different machine learning method. Also using feature reduction, feature number reduces to 10 and classifed with 4 different machine learning method. Performance of current system, developed system with 76 feature data set and reduced 10 feature data set is compared. Speech, picture and hand writing recognition researches using deep learning methods are very popular and successful. Using the same data set Machine learning methods and deep learning methods performances are compared.