Integrated Anti-Fraud System (project overview)
The research results of Integrated Anti-Fraud System (IAFS) project will be presented through a special workshop as a part of the SoftCOM 2019 conference. The IAFS project is a cooperation of university and ICT industry, namely the cooperation between University of Zagreb Faculty of Electrical Engineering and Computing and Multicom. This research has been supported under Competitiveness and Cohesion Operational Programme from the European Regional and Development Fund (project no. KK.01.2.1.01.0041).
IAFS project focused on research and development of fraud detection methods over endless data streams in telecommunication and financial industries. Fraudulent activities are generating big financial losses in the industry today. For the telecommunication industry, Communications Fraud Control Association estimates global fraud loss in 2015 at 38.1 billion US dollars. In 2017, this loss is estimated slightly below 30 billion US dollars. Such numbers create an imperative for detecting and preventing fraudulent activities. Research on the project included semi-supervised methods for anomaly detection in data streams. The first part of the research activities focused on data object unsupervised classification methods, e.g., data stream clustering algorithms. Majority of data stream clustering algorithms work in two distinct phases, separating classification and model updating. A new statistical single-phase data stream clustering algorithm that unifies classification and model updating activities under the single phase was proposed and developed by the project research team. In the second part of the project, research activities switched to methods for detecting data object sequences in data streams, focusing mainly on the information theory, data compression, minimal description length (MDL) theory, and Kolmogorov complexity. In the compression theory context, special attention was given to finite state machines used in data compression and usage of automata to detect regularly occurring data object sequences in data streams. An extension of pushdown automata (PDA) was proposed to cover capturing and detection of multi-contextual data object sequences. Capturing frequency and statistics on transitions, the proposed automata is capable of detecting whether the classified data object sequence is occurring regularly or sparsely. To prevent automata overfitting and complexity explosion, additional methods for the proposed automata compression and transformation were developed. Future research in the area will focus on further advancement and testing of the proposed anomaly detection mechanism based on the single-phase data stream clustering algorithm combined with the extended pushdown automata.
Boris Vrdoljak is full professor at the University of Zagreb, Faculty of Electrical Engineering and Computing. He received his Ph.D. degree from the same faculty in 2004. He spent 3 months as a visiting researcher at the University of Bologna, Italy, and 12 months as postdoctoral researcher at INRIA institute, France. His research interests cover ontology matching, e-business security, data warehousing, and big data analytics. He is manager of the Faculty of Electrical Engineering and Computing team in the project Integrated Anti-Fraud System (IAFS). Boris Vrdoljak is a member of the Centre of Research Excellence for Data Science and Advanced Cooperative Systems (ACROSS-DataScience), Data Streams Laboratory, and Laboratory for Information Security and Privacy. He is also president of the council of the postgraduate specialist study Information Security.