CLUSTER ANALYSIS METHOD FOR PERSONALITY IDENTIFICATION BASED ON EEG DATA

Abstract

In this work, we propose a method for personality identification based on the analysis of electroencephalographic (EEG) data using cluster analysis techniques. The method is grounded on the hypothesis that EEG signals contain biometric information specific to each individual. Our approach involves capturing EEG data under controlled conditions of rest and minimal external interference, allowing us to extract unique patterns from each subject's brain activity.
The experimental setup utilizes a 22-channel EEG recording device, with repeated signal acquisitions taken from two test subjects under identical conditions. Each subject undergoes multiple testing sessions with repeated electrode placements to ensure signal variability and consistency. From the EEG recordings, we extract amplitude tables, power spectrum data, and frequency distributions, which serve as the foundation for statistical and cluster analysis.
The clustering algorithm centers around the calculation of barycenters for each EEG channel. These barycenters represent average amplitude values and serve as a statistical signature of a subject’s EEG profile. We use the Euclidean distance to compare new (control) EEG signals to these established clusters. A smaller distance indicates a higher likelihood of match, confirming the EEG's identity correspondence.
Data processing and analysis are carried out using tools such as MathCad and Excel, allowing for efficient manipulation of EEG matrices and extraction of meaningful metrics. Clusters of amplitude, power, and frequency are formed for each test subject. The verification process involves comparing control EEG signals to these clusters, evaluating how closely the new data aligns with the established barycenters.
Results demonstrate that the control EEG of subject “Andre” deviates 49% from their barycenter, while it deviates 54% from “Vlad's” barycenter, confirming Andre’s identity. Similarly, Vlad’s EEG is closer to his own barycenter than to Andre's. The same identification accuracy is observed across power and frequency clusters.
This cluster-based approach proves to be a reliable method for biometric identification using EEG. The presented method does not require complex preprocessing or machine learning models, relying instead on statistical consistency and barycenter proximity. Our findings support the viability of EEG data as a biometric identifier and highlight the potential for real-world application in identity verification systems.
Keywords: electroencephalogram, cluster analysis, biometric identification, statistical analysis, barycenter signature, emotional state, resting state