EEG-Based Tool for Prediction of University Students’ Cognitive Performance in the Classroom

Figures

• 

  Figure 1. Proposed methodology for implementation of a cognitive performance predictive tool. EEG recordings were measured for text and video groups during baseline (EC, EO), learning tasks and evaluation for three repetitions. EEG data were transferred to a PC via OpenVibe and analyzed using Matlab to build the predictive models.

• 

  Figure 2. Representative EEG recordings during the first five seconds of eyes open, eyes closed and learning tasks for text (top) and video (bottom) groups, measured at electrodes: FP1, FP2, C3, C4, P7, P8, O1 and O2. The spatial distribution of EEG electrodes is shown in the top left inset. Topographic plots show the average α/θ ratios (normalized between −1.5 and 1.5), computed from each five-second EEG recording for all electrodes. The presented EEG signals correspond to those of participants Pt1 and Pv4 for the text and video groups, respectively, during the first LT repetition.

• 

  Figure 3. Flowchart of the experimental protocol implemented in this study. Participants were invited to the study and, after providing consent, performed the learning tasks. After EEG signals were measured and analyzed, relevant features were used to build and evaluate models.

• 

  Figure 4. Average scores (top) and exam times (bottom) for text and video learning trials for all three repetitions. Black lines represent standard deviation.

• 

  Figure 5. Channels and frequency bands which showed statistically significant (p<0.05) higher power during video compared to text (V > T), lower power during video compared to text (V < T) and no significant changes between groups (V = T). Three colormaps represent the three LT repetitions: 1 (top), 2 (middle) and 3 (bottom).

• 

  Figure 6. Channels and power ratios which showed a statistically significant (p<0.05) higher power difference during video compared to text (V > T), lower power difference during video compared to text (V < T) and no significant changes between groups (V = T). Three colormaps represent the three LT repetitions: 1 (top), 2 (middle) and 3 (bottom).

• 

  Figure 7. Correlation between C3 (θ/α) ratio and test scores for all participants from both video and text groups, at all LT repetitions.

• 

  Figure 8. Correlation between normalized FP1 (δ) power and test scores for all participants from both video and text groups, at all LT repetitions.

• 

  Figure 9. Average accuracy between real and predicted scores for MLR models using 1 to 4 features of EEG data.

• 

  Figure 10. Average real and predicted scores for text and video groups across cross-validations. Black lines represent standard deviation.

• 

  Figure 11. Boxplots showing the distribution of (a) ETs, (b) scores and (c) C3 (α/θ) across multiple factors: group (text and video), difficulty (repetitions) and both.