Use of Machine Learning Methods for Solving Problem of User Identifying by Keyboard Handwriting

Abstract

This article deals with the problem of ensuring protection against unauthorized access to data by identifying users according to biometric characteristics. The authors consider keyboard typing, which is a set of characteristics of the user on the keyboard. During the experiments were obtained handwriting samples of several users who printed phrases consisting of all letters of the alphabet. As a result, a statistical sample was obtained for each participant of the experiment, and it allowed concluding that the handwriting samples of each person are stable and sufficient for identification. Keyboard handwriting is a dynamic (changing) characteristic that can be described by many parameters, so the choice of data processing is a difficult task. To analyze the data and identify the author of the handwriting sample, the article provides an overview of existing methods and discusses statistical methods such as k-means and regression analyses, which are the most commonly, used statistical classification methods. Evaluation of the effectiveness of the results is evaluated using the coefficients of false access (FAR) and false denial of access (FRR). The results suggest that these methods do not effectively solve the problem, so it is necessary to use other methods. Thus, the authors proposed the use of artificial intelligence methods that allow us to find hidden patterns and dependencies in the data stream.