Baklan I. Informational technologies based on Bayesian networks for a user recognition

The thesis deals with the problem of biometric pattern recognition by development of a new method for construction and application of discrete Bayesian networks (BN) and dynamic Bayesian networks (DBN), methods of generation of probabilistic inference, methods of training and application of BN and DBN for a pattern recognition problem solution, methods of trajectory data processing and creating on their basis of the new user recognition informational technology. A review and analysis of methods for biometric pattern recognition and substantiation of efficiency of BM vehicle application is considered. To construct BN structure a heuristic learning procedure is developed that uses statistical data for training. As a scoring measure for determining the model structure the minimum recognition error measure is used (MRE). It also has been proposed the method for DBN training using statistical data. The new method of processing trajectory data of cursor movements is proposed. It's based on principal components method, which allows determining of informative features for recognition from a wide array of trajectory data. Numerical experiments evaluate the accuracy of the developed user recognition technology at 92%, which is 5-7% higher than accuracy of a similar approach using the neural networks. An original user recognition information technology has been developed and implemented on the basis of BN. The system is based on the proposed methods of BN structure training, probabilistic inference generation and trajectory data procession. The technology is been used at the "IKS-Techno" company and "BMS-Consulting" company. The scientific results acquired are also been used in several student courses at the NTUU "KPI".