Brain-activity based Machine Learning Models predict three-class Cognitive Performance during Multimodal and Traditional Learning Academic Article in Scopus uri icon

abstract

  • The integration of new technologies and data analysis methods ushered in innovative teaching practices that depart from conventional methods. The advent of information technologies and the rapid adoption of dynamic teaching tools have transformed classrooms into interactive spaces where students and educators engage in real-time feedback and collaborative learning experiences. This study employs electroencephalography (EEG) and machine learning to evaluate the effectiveness of various teaching modalities on students' cognitive performance. EEG, a non-invasive technique that measures brain activity through electrical signals, offers insights into cognitive processes by analyzing event-related potentials (ERPs) and frequency-domain features. The Power Spectral Density (PSD) and Fast Fourier Transform (FFT) techniques are utilized to analyze EEG signals and identify patterns of brain activity in response to different learning environments. The study encompasses EEG data acquisition, pre-processing, and feature extraction, particularly focusing on the Task Engagement Index (TEI) as a measure of cognitive workload and engagement. Employing machine learning algorithms, such as the Multilayer Perceptron (MLP), the proposed algorithm is able to predict students' cognitive performance based on EEG features, categorized in three classes of increasing performance (score). Frequency domain analysis reveals distinct patterns of brain activity within different learning contexts, while the model evaluation underscores the efficacy of the MLP algorithm in predicting cognitive performance; and correlation analysis highlights specific EEG features that significantly influence cognitive performance. The analysis included frequency domain assessment, revealing distinct power patterns between text and video learning groups. MLP algorithm outperformed other models in accuracy, precision, recall, and F1 score evaluations. Statistical analysis highlighted correlations between EEG features and cognitive performance. Results showed similar performance for Text (around 80%) and Video (around 86%) (three-class) predictive models. Confusion matrices confirmed accurate classification. Moreover, models were tested with twenty random samples from each group, obtaining 90% and 95% accuracies for Video and Text respectively. In addition, The study identified cognitive load variations, showcased MLP's effectiveness, and emphasized EEG feature correlations in predicting cognitive performance. © 2023 IEEE.

publication date

  • January 1, 2023