TY - JOUR
T1 - Meta-Heuristic Optimization and Keystroke Dynamics for Authentication of Smartphone Users
AU - El-Kenawy, El Sayed M.
AU - Mirjalili, Seyedali
AU - Abdelhamid, Abdelaziz A.
AU - Ibrahim, Abdelhameed
AU - Khodadadi, Nima
AU - Eid, Marwa M.
N1 - Publisher Copyright:
© 2022 by the authors.
PY - 2022/8
Y1 - 2022/8
N2 - Personal Identification Numbers (PIN) and unlock patterns are two of the most often used smartphone authentication mechanisms. Because PINs have just four or six characters, they are subject to shoulder-surfing attacks and are not as secure as other authentication techniques. Biometric authentication methods, such as fingerprint, face, or iris, are now being studied in a variety of ways. The security of such biometric authentication is based on PIN-based authentication as a backup when the maximum defined number of authentication failures is surpassed during the authentication process. Keystroke-dynamics-based authentication has been studied to circumvent this limitation, in which users were categorized by evaluating their typing patterns as they input their PIN. A broad variety of approaches have been proposed to improve the capacity of PIN entry systems to discriminate between normal and abnormal users based on a user’s typing pattern. To improve the accuracy of user discrimination using keystroke dynamics, we propose a novel approach for improving the parameters of a Bidirectional Recurrent Neural Network (BRNN) used in classifying users’ keystrokes. The proposed approach is based on a significant modification to the Dipper Throated Optimization (DTO) algorithm by employing three search leaders to improve the exploration process of the optimization algorithm. To assess the effectiveness of the proposed approach, two datasets containing keystroke dynamics were included in the conducted experiments. In addition, we propose a feature selection algorithm for selecting the proper features that enable better user classification. The proposed algorithms are compared to other optimization methods in the literature, and the results showed the superiority of the proposed algorithms. Moreover, a statistical analysis is performed to measure the stability and significance of the proposed methods, and the results confirmed the expected findings. The best classification accuracy achieved by the proposed optimized BRNN is 99.02% and 99.32% for the two datasets.
AB - Personal Identification Numbers (PIN) and unlock patterns are two of the most often used smartphone authentication mechanisms. Because PINs have just four or six characters, they are subject to shoulder-surfing attacks and are not as secure as other authentication techniques. Biometric authentication methods, such as fingerprint, face, or iris, are now being studied in a variety of ways. The security of such biometric authentication is based on PIN-based authentication as a backup when the maximum defined number of authentication failures is surpassed during the authentication process. Keystroke-dynamics-based authentication has been studied to circumvent this limitation, in which users were categorized by evaluating their typing patterns as they input their PIN. A broad variety of approaches have been proposed to improve the capacity of PIN entry systems to discriminate between normal and abnormal users based on a user’s typing pattern. To improve the accuracy of user discrimination using keystroke dynamics, we propose a novel approach for improving the parameters of a Bidirectional Recurrent Neural Network (BRNN) used in classifying users’ keystrokes. The proposed approach is based on a significant modification to the Dipper Throated Optimization (DTO) algorithm by employing three search leaders to improve the exploration process of the optimization algorithm. To assess the effectiveness of the proposed approach, two datasets containing keystroke dynamics were included in the conducted experiments. In addition, we propose a feature selection algorithm for selecting the proper features that enable better user classification. The proposed algorithms are compared to other optimization methods in the literature, and the results showed the superiority of the proposed algorithms. Moreover, a statistical analysis is performed to measure the stability and significance of the proposed methods, and the results confirmed the expected findings. The best classification accuracy achieved by the proposed optimized BRNN is 99.02% and 99.32% for the two datasets.
KW - authentication
KW - Bidirectional Recurrent Neural Network
KW - Dipper Throated Optimization
KW - feature selection
KW - keystroke dynamics
KW - meta-heuristic optimization
KW - smartphone
UR - http://www.scopus.com/inward/record.url?scp=85137397588&partnerID=8YFLogxK
U2 - 10.3390/math10162912
DO - 10.3390/math10162912
M3 - Article
AN - SCOPUS:85137397588
VL - 10
JO - Mathematics
JF - Mathematics
SN - 2227-7390
IS - 16
M1 - 2912
ER -