On the Design and Analysis of a Biometric Authentication System Using Keystroke Dynamics

Abstract

this paper proposes a portable hardware token for user's authentication; it is based on the use of keystroke dynamics to verify users biometrically. The proposed approach allows for a multifactor authentication scheme, in which a user cannot be granted access unless they provide a correct password on a hardware token and their biometric signature. The latter is extracted while the user is typing their password. This paper explains the design rationale of the proposed system and provides a comprehensive insight in the development of a hardware prototype of the same. The paper also presents a feasibility study that included a systematic analysis based on training data obtained from 32 users. Our results show that dynamic keystroke can be employed to construct a cost-efficient solution for biometric user authentication with an average error rate of 4.5%

A Secure and Scalable Behavioral Dynamics Authentication Model

“…In [18] author developed a unique way to perform KD-based authentication with a keyboard and an array of pressure sensors that serves to develop unique user profiles that improve the suggested system's efficiency and, using a realworld dataset, achieved a 97% success rate in experiments. Using a natural language processing method, [56] introduced a portable hardware token for MFA using keystrokes to enhance authentication, but the proposed algorithm, though simple, achieved relatively low accuracy, but with a relatively high error rate. So, they suggested applying ML and considering close keystrokes to reduce authentication errors.…”

Fixed-Text vs. Free-Text Keystroke Dynamics for User Authentication

“…Results included an EER of 4.9 with majority voting (MV) considered for selecting specific features and an EER of 6.6 when all features were considered. Robert Cockell and Basel Halak [11] found two statistics-based ways for data analysis. The first was based on the simple averages computation while the second was based on a probability estimation, with both ways depending on the characteristics extracted from each button press.…”

“…Results of using different classifiers with the free-text dataset[11] Method TP Rate (%) FP Rate (%) Average Roc Area (%) Accuracy (%)…”

“…Results of applying different combinations of classifiers with the combination rules Average of Probabilities (AoP) and Major Voting (MV) on the free-text dataset[11] Despite having the highest achieved accuracy (81.70 %) after combining the classifiers in Table9less than before the combination, it is still higher than using 3 individual classifiers results in Table8The Highest Accuracy Rate was 87.58 using Random Forest classifier We did not make a comparison table for the free-text dataset results because it is a new dataset and no published work done on it up till now.…”