Analysis of the impact of online education using EEG signals and machine learning algorithms

Abstract

Online learning has allowed students from different walks of life to access a vast amount of information, allowing them to gain new skills. However, only having access to that information does not mean that the students will comprehend it. In this report, we study the impact of online education on students, specifically their confusion levels. The dataset that we have used in this report was taken from Kaggle. The dataset consists of mostly preprocessed Electroencephalogram (EEG) brain wave values i.e., Attention, Mediation, Raw, Delta, Theta, Alpha, Beta, and Gamma. Due to the limitations of the dataset, the accuracies of the Machine Learning models when only using EEG signal values were not satisfactory. Therefore, later into our research, we have decided to modify our dataset in order to better determine the confusion level of students. We have synthesized the dataset taken from Kaggle to form another dataset, where we took the content being viewed into account which led to better classification. The Machine Learning Algorithms that we have implemented in this paper are Decision Tree, Random Forest, Bagging with Random Forest, Gaussian Naive Bayes, K-Nearest Neighbors, Gradient Boosting, XGBoost, and Bidirectional-LSTM. For the dataset which consists of only EEG signal values, Bagging with Random Forest algorithm performed the best. It was able to predict whether or not a student was confused with an accuracy of 67.3%, while in the modified dataset, Bidirectional-LSTM had the highest accuracy of 80.9%. For both of the datasets, Gaussian Naive Bayes performed the worst with an accuracy of 59.2% and 63.6%, respectively.