Brain Activity Monitoring Using Machine Learning and Dry-EEG Signals

This paper presents the development and testing of a novel brain monitoring system utilizing machine learning and dry-EEG technology. The system employs the wearable DSI-24 sensor and a 1D Convolutional Neural Network (1D CNN) to analyze and interpret signals, classifying data into stressed and calm driving states. The study evaluates the model's performance under balanced and unbalanced class distributions. Results indicate that balanced datasets yield higher precision, recall, and F1 scores, with balanced classes achieving an F1 score of 83.3% in validation and 80% in testing, compared to 85.7% and 73.7% for unbalanced classes. Addressing class imbalance through techniques like oversampling, undersampling, and synthetic data generation is recommended to enhance model reliability. The developed model, when optimized for class balance, is expected to provide more accurate and trustworthy predictions, thereby better generalizing to unseen data. These insights offer valuable guidelines for future model development and deployment, ensuring robust performance across varying data distributions.