A Machine Learning-Based Approach to Password Authentication Using Keystroke Biometrics

Abstract

Keystroke authentication systems, though they are getting more popular, are a rather common method of data/network access. The typing dynamics apply to the automatic process of recognizing or verifying an individual’s identity depending on the form and style of tapping on a keyboard. It allows to authenticate individuals through their way of typing their password or a free text on a keyboard. In this paper, we perform analysis of machine learning algorithms on a keystroke dynamics-based data set with features like Hold time, Keyup-Keydown time and Keydown-Keydown time. This research is based on our methodology of using support vector machines (polynomial & radial basis kernels), random forest algorithm and artificial neural networks to recognize users based on their keystroke patterns. The review analysis shows a great result in user identification based on keystroke patterns with artificial neural networks showing better results among the three algorithms implemented at 91.8% accuracy.