User Behavior Authentication Based on Computer Mouse Dynamics

The aim of this work was to investigate existing algorithms for dynamic user authentication and develop our own, based on an analysis of computer mouse handling characterized by high-quality performance and a supporting dynamic mode. Existing ways of constructing and preliminarily processing the feature space are considered, along with means of dynamic authentication based on the use of classical techniques of machine learning and neural networks. Modifications with better efficiency are proposed for the ones most promising. A cross-platform application for dynamic user authentication based on an analysis of computer mouse handling is designed and implemented using the proposed technique, which shows the best performance (ROC AUC = 0.82). This system and its individual modules can serve as a basis for building advanced information security systems. Experimental studies are conducted that confirm the validity of the obtained results and the correctness of the system’s operation.