Assessing a Single Student’s Concentration on Learning Platforms: A Machine Learning-Enhanced EEG-Based Framework
1 st Zewen Zhuo Faculty of Science and Technology University Paris-Est Créteil Créteil, France [email protected] 2 nd Mohamad Najafi Faculty of Science and Technology University Paris-Est Créteil Créteil, France [email protected] 3 rd Hazem Zein LISSI Laboratory University Paris-Est · 2025
Read Assessing a Single Student’s Concentration on Learning Platforms: A Machine Learning-Enhanced EEG-Based Framework by 1 st Zewen Zhuo Faculty of Science and Technology University Paris-Est Créteil Créteil, France [email protected] 2 nd Mohamad Najafi Faculty of Science and Technology University Paris-Est Créteil Créteil, France [email protected] 3 rd Hazem Zein LISSI Laboratory University Paris-Est (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-15107.txt.
Introduction
Zewen Zhuo, Mohamad Najafi, Hazem Zein, and Amine Nait-Ali gauge student concentration from brain signals in a 2025 paper. Online learning platforms rarely know if a learner is focused. A personalized model per student could give that insight. The team built a full pipeline around EEG headband data.
Fifty statistical features were drawn from five brain wave bands. Careful selection trimmed inputs for the personal model. A consumer headband with five electrodes recorded the signals. Recordings covered computer and virtual reality learning sessions. A tuned random forest classified the student's concentration state. Test accuracy neared ninety-eight percent in both settings. Hyperparameter tuning contributed to that performance. Protocols for capture and preprocessing are detailed for reuse. The authors pitch personal EEG models for adaptive learning tools.
Catalog record
| CREATOR | 1 st Zewen Zhuo Faculty of Science and Technology University Paris-Est Créteil Créteil, France [email protected] 2 nd Mohamad Najafi Faculty of Science and Technology University Paris-Est Créteil Créteil, France [email protected] 3 rd Hazem Zein LISSI Laboratory University Paris-Est |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 20 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2502-15107
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Assessing a Single Student’s Concentration on Learning Platforms: A Machine Learning-Enhanced EEG-Based Framework Thanks: ———————————- * Corresponding Author 1 st Zewen Zhuo Faculty of Science and Technology University Paris-Est Créteil Créteil, France [email protected] 2 nd Mohamad Najafi Faculty of Science and Technology University Paris-Est Créteil Créteil, France [email protected] 3 rd Hazem Zein LISSI Laboratory University Paris-Est Créteil Vitry-sur-Seine, France [email protected] 4 th * Amine Nait-Ali LISSI Laboratory University Paris-Est Créteil Vitry-sur-Seine, France [email protected] Abstract This study introduces a specialized pipeline designed to classify the concentration state of an individual student during online learning sessions by training a custom-tailored machine learning model. Detailed protocols for acquiring and preprocessing EEG data are outlined, along with the extraction of fifty statistical features from five EEG signal bands: alpha, beta, theta, delta, and gamma. Following feature extraction, a thorough feature selection process was conducted to optimize the data inputs for a personalized analysis. The study also explores the benefits of hyperparameter fine-tuning to enhance the classification accuracy of the student’s concentration state. EEG signals were captured from the student using a Muse headband (Gen 2), equipped with five electrodes (TP9, AF7, AF8, TP10, and a reference electrode NZ), during engagement with educational content on computer-based e-learning platforms. Employing a random forest model customized to the student’s data, we achieved remarkable classification performance, with test accuracies of 97.6% in the computer-based learning setting and 98% in the virtual reality setting. These results underscore the effectiveness of our approach in delivering personalized insights into student concentration during online educational activities. I Introduction In the past few years, e-learning has gained more attention due to the global COVID-19 pandemic, during which online education became the primary method of imparting knowledge. Nowadays, hybrid learning remains a prevalent modality in many universities and institutions because of its benefits such as low cost, unlimited access, and the availability of comprehensive materials. Notably, with the advancement of virtual reality (VR) systems, an increasing number of users have switched their learning format from traditional e-learning devices, like personal computers and tablets, to VR for a more immersive experience. Understanding learners’ mental states during online study sessions can be complicated but is necessary for lecturers to adjust the curriculum and assist students with difficulties [ 1 ] . One effective method for assessing human concentration is electroencephalography (EEG), a noninvasive technique that collects informative brain data via electrodes placed on the scalp. Typically, EEG signals are segmented into five bands based on their frequency range: delta ( \delta ) for frequencies less than 4 Hz, theta ( \theta ) for 4-7 Hz, alpha ( \alpha ) for 8-12 Hz, beta ( \beta ) for 13-30 Hz, and gamma ( \gamma ) for frequencies above 30 Hz [ 2 ] . Although understanding attentional states requires expertise in various fields, including sensor technology and neurophysiology, the rapid advancement of machine learning (ML) technology has facilitated scientists in analyzing these complex signals. The deployment of wearable EEG devices has gained significant popularity among research groups worldwide for classifying cognitive states and other purposes. Classification between concentration and meditation using a random forest classifier [ 3 ] explored different statistical features and manifested an accuracy of 75%. It demonstrates the robustness of utilizing wearable EEG devices and ML models for mental state classification. Mental state recognition in real-time in the e-learning environment [ 4 ] depicted a promising perspective of maximizing users’ learning performance in the adaptive format using advanced ML techniques. Multi-class classification among relaxing, neutral, and concentrating through a combination of features and classification models like Bayesian networks, support vector machines, and random forest [ 5 ] leveraged the overall classification accuracy to 87% and unleashed the potential of mental state classification with higher accuracy by sophisticated feature selection. A major challenge of EEG data classification is extracting meaningful features and mapping them to multiple categories. The current research mainly focuses on the binary classification of EEG signals, which is insufficient for understanding online learning performance. Furthermore, the lack of comparison of model accuracy with different numbers of selected features impedes efforts to maximize model performance. Furthermore, the investigation on VR-based e-learning is still in its early stages, requiring more study to integrate VR in online learning concentration estimation. To address these issues, we propose a pipeline to develop ML models for individual user-based concentration state classification in both VR and non-VR learning environments. It provides a thorough approach starting from data collection to feature selection and model fine-tuning, applying the most effective generic method to analyze the attentional states in different learning formats. This method not only enhances the performance of ML models in multi-class mental state classification but also opens new avenues for comparing different e-learning formats. This paper is structured as follows: the experimental setup, methodology, and specifications regarding the deployed devices are summarized in Section II . Section III manifests the detailed results as well as comparison, performance evaluation, and some further discussions. A final conclusion and other future perspectives are provided in Section IV . Fig. 1 : Muse S EEG Headband : the green and yellow labels correspond to the international 10-20 EEG electrode placement standard II Materials and Methodology II-A EEG Data Acquisition The study aims to provide guidelines for achieving ML-based concentration classification in two different e-learning formats utilizing EEG data. Traditional clinical-level EEG setups are expensive and hard to operate without expertise [ 6 , 7 , 8 ] . In this research, we utilized the Muse S (Gen 2) , as shown in Fig. 1 , which has been employed in diverse research circumstances [ 9 , 10 , 11 ] . This portable and lightweight headband is a commercial EEG sensing equipment, embedded with five sensors. Each sensor is positioned in the specific region as illustrated in Fig. 2 , including TP9, AF7, AF8, and TP10, to record brain activities, along with a reference point (NZ) to calibrate the brain signals. In most clinical settings, EEG signals are recorded as raw data from each electrode. However, these signals are typically intricate and challenging to interpret without the assistance of specialists. One effective way to simplify the analysis is to decompose the raw EEG signals into five distinct frequency bands through the application of band-pass filters [ 12 ] . Besides that, many studies stated that the power line interference can significantly undermine EEG signal analysis [ 13 , 14 ] . The common practice is to apply a notch filter (50 or 60 Hz, depending on the geographic region) to phase out undesired power noise, thereby mitigating signal contamination. The availability of Mind Monitor on various mobile devices and its capacity to apply notch and band-pass filters made it ideal for data acquisition. Furthermore, it can record the processed signals. During data collection, a sampling frequency of 256 Hz was selected to ensure the capture of meaningful features for the classification task. Fig. 2 : The International 10-20 EEG Electrode Placement Standard [ 5 ] : the yellow labels are electrodes of Muse headband and green labeled sensor is deployed as a reference point for calibration Aiming to provide generic model development pipelines for both conventional e-learning (e.g. on personal computers) and VR-based