University of the Aegean
Behavioral biometrics for continuous authentication: security and privacy issues
Abstract
dc:descriptionAuthentication systems based on PINs and passwords or physiological biometrics (e.g., iris, fingerprint, etc.) establish the user identity only at the beginning of the session using the entry-point authentication model. This model has been criticized heavily for being vulnerable to attacks occurring after the authenticated session has been established. Some of these systems defend against such attacks by performing an additional authentication step at critical points of the session but are unpopular with users due to repetitive authentication. Mobile devices and their applications use the entry-point authentication model to authenticate users. Therefore, an important concern is to determine whether the mobile device is in the hands of the genuine user, and accordingly, whether the genuine user is the one using the sensitive services during a session. To solve this issue Continuous Authentication (CA) systems using Behavioral Biometrics (BBs) have been proposed in the literature. CA systems are an additional security measure that monitors users' biometric behavior by constantly re-authenticating them during a login session. However, their practical application is limited due to two fundamental shortcomings. The first shortcoming focuses on non-technical issues, for example, perceptions related to the fears and expectations of future users and perceived concerns about the privacy of their biometrics. The second shortcoming focuses on the problem of false positive/false negative results, that is, on security and usability issues.This doctoral thesis includes four research stages. These are the research stages of a single research project. In the first research stage, an extensive literature review is presented that maps the research area and concerns BBCA technology and the performance of machine learning systems. Additionally, a literature review on potential attack vectors on BBCA systems is presented, and promising countermeasures are highlighted. Also, behavioral biometrics (Behavioral Biometrics - BBs) are classified into seven categories and an analysis of their feature extraction and collection methodologies is carried out. Finally, challenges, open issues, and future trends are identified.In the second research stage of this thesis, the effect of various factors on Behavioral Intention to Adopt the Technology (Behavioral Intention - BI) is investigated through a new integration of the Technology Acceptance Model (TAM) and the Diffusion of Innovation Theory (DOI). Also, a new theoretical framework is developed with constructs such as Security & Privacy Risks (SPR), Biometrics Privacy Concerns (BPC), and Perceived Risk of Technology Use (PROU). In addition, the constructs of Trust in Technology (TT) and Innovativeness (Innov) were used.It was found that the main Facilitators of the Behavioral Intention to Adopt the Technology (BI) are Trust in Technology (TT), followed by Compatibility (COMP), Perceived Usefulness (PU), and Innovativeness (INNOV). This research also shows that people care less about the ease of use of technology and are willing to sacrifice it to achieve greater security. Compatibility and Innovation also play an important role. People who believe that using BBCA technology would fit their lifestyle and would like to experiment with new technologies have a positive intention to adopt BBCA technology. For the new constructs added, the results support the hypothesis that Security & Privacy Risks (SPR) is a facilitator of Perceived Usefulness (PU). Also, PU acts as a facilitator of Behavioral Intention to adopt the technology (Behavioral Intention – BI). Consequently, individuals who do not feel sufficiently protected by classic authentication methods will consider the usefulness of BBCA technology for their additional protection against risks. Also, the constructs Biometrics Privacy Concerns (BPC) and Perceived Risk of Using the technology (PROU) examine whether individuals' concerns about their biometric privacy act as inhibitors to BI. The conclusion drawn is that individuals consider the benefits related to the security of their assets (e.g., money in a bank account) from using BBCA technology to be far more important than the perceived risks to the privacy of their biometrics. This results from the fact that the hypothesis that the main inhibitor of BI is the construct Perceived Risk of Using the technology (PROU) is not supported by the model. The new constructs were used to extend the TAM model and address its limitations in addressing perceived security and privacy issues. Therefore, it is proposed that the new theoretical framework be combined with TAM for research on the adoption of biometric authentication and continuous authentication technologies.In the third research stage of this thesis, it was observed that a major challenge and an open problem for research related to the development of BBCA systems, is the collection and refinement of an appropriate set of behavioral biometric data. The issue is compounded by the fact that most users avoid engaging in time-consuming and laborious processes involved in collecting biometric survey data. For this reason, developing and testing a user-friendly biometric collection methodology and tool is another major challenge. In response to these challenges, in the third stage of the research, a new behavioral biometric data collection paradigm, called the BioGames paradigm, is presented and follows an innovative approach. This approach is about gamification of data collection. At the same time, a behavioral biometric collection tool (Biogames App) based on the BioGames example was developed. The BioGames App uses games and challenges that combine Keystroke Dynamics and Touch Gestures.In the fourth stage, research related to the design and evaluation of new approaches to continuous authentication using Keystroke Dynamics and Touch Gestures is presented. According to the literature, several studies use single behavioral modality methods to authenticate users. However, the behaviors of genuine users may change, and systems fail when significant changes occur. The above result in either security or usability issues. In the literature, Biometric Fusion is used to solve this problem and achieves improved results. In the present research, each behavioral biometric case is examined separately and the case of improving performance results by fusion of touch gestures and keystroke dynamics at the feature level (Feature-level fusion) is investigated. In the present approach, a comparison is made between deep neural networks designed for data that entail important temporal dynamics, such as Multi-Layer Perceptron (MLP), and deep networks designed for independently distributed data, such as Long Short-Term Memory (LSTM). By comparing the performance of both systems, the MLP is superior to LSTM in this context. The MLP achieved greater improvement and better performance compared to the LSTM. The MLP achieved an Accuracy of 98.3% (an increase of 21.1%), an EER of 1% (the error was reduced by 23.7%), a TAR of 99.4% (an increase of 46%), a TRR of 97.4% (increased 10%), a FAR of 2.6% (reduced by 10.5%) and an FRR of 0.6% (reduced by 46%). The research results show that the feature-level fusion of touch gestures and keystroke dynamics improves the performance of the systems and solves both security and usability issues.
Degree
thesis:*- Grantor dc:publisher
- University of the Aegean
- Year dc:date
- 2023
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Stylios, Ioannis
- Στύλιος, Ιωάννης
Subjects
dc:subject × 22- Βιομετρικά συστήματα
- Αυθεντικοποίηση χρήστη
- Ασφάλεια πληροφοριακών και επικοινωνιακών συστημάτων
- Ασφάλεια κινητών τηλεφώνων
- Μηχανική μάθηση
- Βαθιά μάθηση
- Μοντέλα αποδοχής της τεχνολογίας
- Biometric systems
- User authentication
- Information and communication systems security
- Mobile security
- Machine learning
- Deep learning
- Technology acceptance models
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Φυσικές Επιστήμες
- Τεχνητή νοημοσύνη
- Επιστήμη ηλεκτρονικών υπολογιστών και Πληροφορική, άλλοι τομείς
- Computer and Information Sciences
- Natural Sciences
- Artificial Intelligence
- Computer and Information sciences, miscellaneous
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/53367