Using EEG Brain Signals to Predict Children’s Learning Performance During Technology-assisted Language Learning

Abstract

Brain-computer interface (BCI) systems can provide an objective evaluation of a learner’s performance based on their brain signals. Having an accurate assessment of the learning process can facilitate the adaptation of the educational setting to improve the effectiveness of learning. The current study aimed at developing BCI classifiers that can accurately predict the learning performance of children by investigating temporal and spectral characteristics of their brain signals. EEG brain activity collected from children while learning a second language was analyzed by extracting two types of features: frequency-domain features using power spectral density (PSD) analysis and time-domain features using detrended fluctuation analysis (DFA). Two machine learning models (i.e., SVM and KNN) were trained with either PSD features or both PSD and DFA features. It was expected that adding a time-domain feature would improve the prediction accuracy of the models. However, the models with only frequency-domain features performed significantly better, suggesting that EEG temporal correlations may not reflect learning processes. Despite this, both approaches yielded high prediction performance, with SVM reaching 84% accuracy in predicting the learning performance of children. These findings highlight the potential of neurotechnology to be employed in learning analytics to refine traditional methods of academic performance measurement and ultimately achieve personalized learning.