Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Support Vector Machine for Person Classification Using the EEG Signals

Naveenkumar G Venkataswamy, Masudul H Imtiaz

TL;DR

This study extracts salient features from the EEG signals and trains a supervised multiclass Support Vector Machine classifier, which highlights the viability of machine learning in implementing real-world, EEG-based biometric identification systems, thereby advancing user authentication technology.

Abstract

User authentication is a pivotal element in security systems. Conventional methods including passwords, personal identification numbers, and identification tags are increasingly vulnerable to cyber-attacks. This paper suggests a paradigm shift towards biometric identification technology that leverages unique physiological or behavioral characteristics for user authenticity verification. Nevertheless, biometric solutions like fingerprints, iris patterns, facial and voice recognition are also susceptible to forgery and deception. We propose using Electroencephalogram (EEG) signals for individual identification to address this challenge. Derived from unique brain activities, these signals offer promising authentication potential and provide a novel means for liveness detection, thereby mitigating spoofing attacks. This study employs a public dataset initially compiled for fatigue analysis, featuring EEG data from 12 subjects recorded via an eight-channel OpenBCI helmet. This dataset extracts salient features from the EEG signals and trains a supervised multiclass Support Vector Machine classifier. Upon evaluation, the classifier model achieves a maximum accuracy of 92.9\%, leveraging ten features from each channel. Collectively, these findings highlight the viability of machine learning in implementing real-world, EEG-based biometric identification systems, thereby advancing user authentication technology.

Support Vector Machine for Person Classification Using the EEG Signals

TL;DR

This study extracts salient features from the EEG signals and trains a supervised multiclass Support Vector Machine classifier, which highlights the viability of machine learning in implementing real-world, EEG-based biometric identification systems, thereby advancing user authentication technology.

Abstract

User authentication is a pivotal element in security systems. Conventional methods including passwords, personal identification numbers, and identification tags are increasingly vulnerable to cyber-attacks. This paper suggests a paradigm shift towards biometric identification technology that leverages unique physiological or behavioral characteristics for user authenticity verification. Nevertheless, biometric solutions like fingerprints, iris patterns, facial and voice recognition are also susceptible to forgery and deception. We propose using Electroencephalogram (EEG) signals for individual identification to address this challenge. Derived from unique brain activities, these signals offer promising authentication potential and provide a novel means for liveness detection, thereby mitigating spoofing attacks. This study employs a public dataset initially compiled for fatigue analysis, featuring EEG data from 12 subjects recorded via an eight-channel OpenBCI helmet. This dataset extracts salient features from the EEG signals and trains a supervised multiclass Support Vector Machine classifier. Upon evaluation, the classifier model achieves a maximum accuracy of 92.9\%, leveraging ten features from each channel. Collectively, these findings highlight the viability of machine learning in implementing real-world, EEG-based biometric identification systems, thereby advancing user authentication technology.

Paper Structure

This paper contains 13 sections, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Materials and Methods
  3. Datasets
  4. Experimental Design
  5. EEG Signal Acquisition
  6. Methods
  7. Filtering
  8. Windowing
  9. Feature Extraction
  10. Dimensionality Reduction
  11. Classifier Implementation
  12. Results and Discussion
  13. Conclusions

Figures (3)

  • Figure 1: The design concept for EEG signal acquisition experiments. Biometric data were collected using OpenBCI in the relaxed state (E.C. and E.O.).) and the A.O. task. In the A.O. tasks, frequent and infrequent stimuli are presented randomly in an 80:20 ratio. The data is transferred to the P.C. via Bluetooth with the OpenViBE software ramirez2021evaluation.
  • Figure 2: Grand average representation of the P300 wave across all the participants. Eight traces (one per channel) are presented for the frontal(F), center(C), parietal (P), and occipital(O) electrodes of the left and right hemispheres of the brain. The shaded area represents standard error across traces ramirez2021evaluation
  • Figure 3: Variance plot from PCA. The X-axis shows the percentage variance explained by every 27 principal components. The Y-axis shows the cumulative sum.