A Machine Learning Approach in EEG-Based Assessment of Cognitive Load by Mental Arithmetic Tasks

Abstract In this study, we investigate the characteristics of electroencephalogram (EEG) signals during mental arithmetic tasks. The publicly available dataset used in this study includes EEG recordings from 36 healthy volunteers performing serial subtraction tasks, with data collected using a 23-channel Neurocom EEG system. The study aims to explore the changes in brain activity during cognitive workload by analyzing the power spectral density, entropy, and statistical features of the EEG signals. Key features such as band power in delta, theta, alpha, beta, and gamma frequency bands, entropy, mean power, standard deviation, kurtosis, skewness, autocorrelation, and RMS energy were extracted from each channel. The subjects were categorized into good and bad counters based on their arithmetic performance. The extracted features were then used to train machine learning models, including Linear Discriminant Analysis (LDA), k-Nearest Neighbors (KNN), and Support Vector Machine (SVM), to classify the cognitive workload levels. The KNN model demonstrated the highest overall performance with an F1 score of 0.857. This research highlights the potential of using EEG features to assess cognitive workload and provides a foundation for future studies on brain-computer interfaces and cognitive neuroscience.