Machine Learning Methods for Electroencephalogram-Based Age Prediction

Electroencephalography (EEG) is a technique that measures brain activity across the surface using electrodes attached to the scalp. The advantages of EEG include its noninvasiveness and cost-effectiveness compared to alternatives, like magnetic resonance imaging and electrocorticography. EEG analysis has various applications, such as in detecting neurodegenerative diseases (eg. epilepsy) and facilitating brain-computer interfaces. In this study, we utilize machine learning to explore age prediction based on EEG data. To accomplish this, we employ EEGNet, a convolutional neural network extensively validated in diverse EEG analysis tasks and brain-computer interfaces. Additionally, we use EEGNet as a feature extractor to train classical machine learning models like Random Forest. Furthermore, we train a recurrent neural network to compare the convolutional approach of EEGNet with other deep learning methods. Our research draws on a unique and high-quality dataset collected from multiple hospitals in British Columbia, Canada. Through model training and testing on the diversified dataset, the proposed models prove to have great potential in predicting brain age, an important biomarker for assessing brain health.