Inconsistent electroencephalogram tasks for continuous person authentication using radial basis function-based support vector machine

Wong, Rui Zhen (2022) Inconsistent electroencephalogram tasks for continuous person authentication using radial basis function-based support vector machine. Masters thesis, Universiti Teknikal Malaysia Melaka.

	Text (24 Pages) Inconsistent electroencephalogram tasks for continuous person authentication using radial basis function-based support vector machine.pdf - Submitted Version Download (645kB)
	Text (Full Text) Inconsistent electroencephalogram tasks for continuous person authentication using radial basis function-based support vector machine.pdf - Submitted Version Restricted to Registered users only Download (6MB)

Abstract

Continuous authentication (CA) provides higher security by repetitively verifying the user identity throughout the entire active session. Electroencephalogram (EEG) person authentication is task sensitive, thus the current data acquisition protocol does not advocate frequent re-authentication. Limited evidence has shown that using the uncontrolled data collection protocol is essential for CA implementation. Current literature presented the task consistent examples such as simulated driving and computer typing tasks. However, many other real-life cases, like computer usage activities, are engaging with various inconsistent tasks. This study proposes a novel use of inconsistent EEG tasks to support continuous person authentication based on uncontrolled changing task condition, to resemble the reallife scenarios. This research has adopted the experimental methodology approach. The inconsistent EEG tasks paradigm was designed for common computer usage tasks. The experimental data were collected from 40 healthy subjects using 20 electrodes EEG head cap in the International 10-20 system. The collected signals were pre-processed and equally segmented to 10 seconds epoch without overlapping. Features extracted from Welch’s estimated Power Spectral Density in low-beta band showed the highest discriminant capability and stability. Principal Component Analysis was employed for dimension reduction to the suggested 90% of variance coverage. The Radial Basis Function (RBF) based Support Vector Machine (SVM) was proposed as the continuous authentication technique. Other popular kernel functions, i.e. the Sorensen, Tanimoto, ANOVA, Log, Linear, Polynomial, and Sigmoid, were used as the benchmarking methods. The proposed model has achieved the highest performance of 91.56% F-Measure. This research has derived some important findings. First, the inconsistent EEG tasks authentication perform equally to the existing consistent paradigm with less restriction in the data collection process. Second, the RBF based SVM technique is suggested for the inconsistent EEG authentication modelling since it was reported robust in many EEG literature, and has achieved comparable results to other top performing kernels in this study. Third, the EEG training data size can be reduced to a minimum of 9 trials (equivalent to 90 seconds or 1.5 minutes) while maintaining its effectiveness in modelling for every subject. In conclusion, the uniqueness of individuals in the inconsistent EEG fulfils the biometrics requirements for person authentication, even though the inconsistent tasks caused more noise due to tasks changing in the iterative authentication process. Furthermore, the intersession and template ageing issues are essential factors in biometrics but not considered in this study. Hence, this research’s future work should further investigate the intersession, template ageing, electrodes reduction, and epoch size issues, to increase the usability of EEG CA in real-life scenarios.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Biometric identification, Authentication, Security systems, Electroencephalography
Subjects:	T Technology > T Technology (General) T Technology > TK Electrical engineering. Electronics Nuclear engineering
Divisions:	Library > Tesis > FTMK
Depositing User:	F Haslinda Harun
Date Deposited:	16 Jan 2024 14:27
Last Modified:	16 Jan 2024 14:27
URI:	http://eprints.utem.edu.my/id/eprint/26976
Statistic Details:	View Download Statistic

Actions (login required)

View Item