An Anomaly Detection System For Subscriber Activities In Private Networks

The constantly-increasing arousal of new cybersecurity threats and malfunctionings on communication networks have highlighted the need for efficient and timely anomaly detection algorithms. The recent developments in the field of machine learning have made possible to automatically find anomalies, frauds, brand new bugs and inconsistencies in the subscribers activities logged on a data server. This thesis describes the collection and processing of the Charging Data Records generated from the subscribers connected to a private network. The work shows the feature selection strategies adopted to extract a dataset with significant information and presents different machine learning models whose results were evaluated on some test data. The investigation has been carried on during a thesis internship at Athonet and the final algorithm has been deployed to one private network of an Athonet's customer to see its performances.