Brainprint authentication model using incremental CNN in dynamic environments

Abstract

Predicting research collaboration potential through Electroencephalogram (EEG) signals presents a significant challenge due to the non-stationary nature of brain activity. Traditional Convolutional Neural Networks (CNNs) are often limited by a static learning paradigm, which fails to adapt to signal drift and evolving cognitive states over time. This paper proposes an incremental CNN-based model designed to update its parameters dynamically as new EEG data becomes available. Unlike standard CNNs that require full retraining, this incremental approach utilizes a continuous learning mechanism to integrate new features while mitigating catastrophic forgetting. The proposed architecture was evaluated through a comparative analysis using test data from 45 healthy subjects. Our results demonstrate that the incremental CNN achieves a classification accuracy exceeding 90%, exhibiting superior robustness in long-term monitoring compared to traditional batch-trained CNNs. These findings suggest that an incremental learning framework is more effective for real-time models.