EEG-Based Biometric Identification Model (Autoencoder + CNN)

This model implements a hybrid architecture combining an Autoencoder for feature extraction and a Convolutional Neural Network (CNN) for classification of EEG signals. It is designed for biometric identification using spectrogram-transformed EEG data.

Model Overview

Input: Spectrograms generated from EEG signals.
Architecture:
- Autoencoder: Compresses high-dimensional spectrogram data into compact latent representations.
- CNN Classifier: Learns patterns from either raw spectrograms or encoded features for classification.
Training Dataset: Public EEG Motor Movement/Imagery Dataset (BCI2000), including signals from 109 subjects across 14 tasks.

Performance

The combined Autoencoder + CNN approach achieves significantly improved classification accuracy compared to baseline CNN-only models, with performance metrics including:

Accuracy: Up to 99.6%
F1 Score: High across all subject classes