Enhancing EEG Analysis for Rapid Brain Activities Detection in Patients

Electroencephalogram (EEG) analysis plays a critical role in diagnosing various neurological conditions by detecting abnormal brain activities. However, the complex nature of EEG signals poses challenges for traditional analysis methods, which often struggle to handle diverse and high-dimensional data effectively. This paper develops a new technique of EEG signal classification using machine learning and deep learning models. Specifically, attention has been given to the CatBoost, ResNet34D, EfficientNetB0, and EfficientNetB2 models. The proposed approach utilized a vast EEG dataset with 11,000 samples of spectrograms, and it conducted an in-depth comparison of these models. Data preprocessing included features developed with statistical measures and the formation of uniform EEG spectrograms. All these models were evaluated in terms of their KL divergence score, defined as a measure of the divergence between the predicted and target distributions. From all the models considered, CatBoost proved to be one with an excellent performance, having had a KL score value at 0.78 compared to that of deep learning models. This work demonstrates the potential of gradient boosting for EEG-activity classification tasks and can be of benefit as a source of insight for interested applications in real-time brain activity monitoring and diagnostics of neurological issues.