See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/329495116

Differentiating Between Visual and Non-Visual Learners Using EEG Power

Spectrum Entropy

Conference Paper · August 2018

DOI: 10.1109/ICIAS.2018.8540571

CITATIONS READS
4 1,625

4 authors:

Soyiba Jawed Hafeez Ullah Amin

Brno University of Technology University of Nottingham
8 PUBLICATIONS   30 CITATIONS    73 PUBLICATIONS   1,647 CITATIONS   

SEE PROFILE SEE PROFILE

Aamir Malik Ibrahima Faye

Brno University of Technology Universiti Teknologi PETRONAS
423 PUBLICATIONS   6,051 CITATIONS    190 PUBLICATIONS   2,380 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Point Cloud Descriptors View project

Spatial Modelling of PM10 Hazard Levels in Malaysia Using Machine Learning Techniques View project

All content following this page was uploaded by Soyiba Jawed on 27 December 2018.

The user has requested enhancement of the downloaded file.

Differentiating between visual and non-visual learners
using EEG power spectrum entropy
Soyiba Jawed, Hafeez Ullah Amin, Aamir Saeed Malik, Ibrahima Faye*
Centre for Intelligent Signal & Imaging Research, Department of Electrical and Electronic Engineering
Universiti Teknologi PETRONAS, Tronoh, Perak, 31750, Malaysia
* [email protected]

Abstract— The purpose of this study is to distinguish the technology, statistics, and computational methods to
visual learners from visual non-learners while learning, having distinguish visual learners from non-visual learners [4]. The
no background knowledge of the contents. The learners are question is Why use EEG the answer is KISS (keep it
distinguished analysing their brain patterns. EEG data were simple and silly).
recorded during learning and memory tasks using 128 Many studies investigate EEG measures during learning.
channels machine from a sample of thirty-four healthy
One of the studies measure EEG for high impulsive and low
university students in two sessions. The students were shown
the animated learning content in video format for eight impulsive individuals during instrumental learning task[5].
minutes. The brain waves were measured during leaning task. EEG power and coherence features are extracted from 30
The study characterizes and distinguish between the visual scalp sites for EEG bands. Here learners as compared to
learners and non-visual learners considering the extracted non- learners had higher value of theta on Centro-parietal
brain patterns. The power spectral entropy features are region and reduced value on frontal region. Increased beta
computed for the recorded EEG and is filtered into alpha and power on temporal region and reduced beta power in
beta sub bands. The most suitable features are selected using Centro-parietal region. Learners also have greater coherence
principal component analysis (PCA). These features are than for frontal-parietal than non-learners. The results highlight
given as an input to the k-nearest neighbour(k-NN) classifier.
the importance of a fronto-parietal region in memory and
Feature classification using k-nearest neighbour has attained
testing accuracy of 96% accuracy for alpha and 95% beta learning. This study outlines the importance of the learning
bands for first learning session and 97% and 94% for second process and how individual differences play an important
learning session. The results show's that the alpha and beta role in learning ability.
power spectral entropy represent distinct and stable EEG Another study predicts the learner’s answers from their
signatures for visual learners and non-visual learners while brain waves and emotional dimensions [6]. Twenty-four
performing the learning tasks. participants participated in the experiment. They were asked
to read thirty-five questions for the first time which they
Keywords—EEG, Feature extraction, k-NN, learning style, were supposed to answer the next day. Measuring the brain
visual learner waves tells the participant emotional state when they know
the answer as compare to when they do not know the
I. INTRODUCTION answer.
Each student has different learning capability independent Some work is also done in increasing attention of normal
of other students. Starting from quick learners to slow college student [7]. Subjects are divided into learners and
learner. Preferences based on a quick understanding of the non-learners using feedback practicing standards. The
subject [1]. In this study we classify visual learners from participant of this study is diagnosed with attention deficit
non-visual learners by analyzing their brain patterns disorder. They are provided with EEG biofeedback to
obtained using Electroencephalography (EEG) when a new enhance beta activity while lowering the alpha and theta
material is presented to them. activity. The activity involves intermediate visual and
Electroencephalography (EEG) large scale robust measure auditory continuous performance test. The results show the
of neocortical dynamic function over. The EEG has significant improvement in learners however non-learners
different frequency bands observed at different brain regions show no improvement.
having different brain responsibilities. The focus of many Apart from these many studies are done on reward and
studies is mainly to use EEG to find the individual learning relationships describing how reward enhance
differences in term of information processing[2]. learning [8] [9-12].
Information processing is a relative term when we talk about In this study, we present a method for distinguish the visual
learning. However, the focus of most of the researchers is learners from visual non-learners based on the EEG signals
cognitive style [3], thinking style and learning style. and employed the PSE as a feature extraction, the PCA as a
Analyzing which they can identify the cognitive load of a feature selection technique, and the k-nearest neighbor (k-
student. The frame work of the existing studies involves the NN) as a machine-learning algorithm for classification of
concepts of educational psychology, neuroscience, two groups. We attempt to explore the brain neuronal
 behavior of the visual learners as compared to non-visual structure”, 2) “Brain anatomy and functions”, 3) “Brain
learners when the new information is presented to the disorder (Alzheimer disease)”. The subjects had verified that
students. they have no background knowledge of the animation’s
The paper is organized as follows: section 2 provides the contents.
material and methods. Section 3 explains the results and Based on the video Total of 20 questions were asked. Each
discussion. Followed by section 4 conclusion. MCQ comprises of missing information with four choices,
out of which one was the correct answer. The subjects had
limited time to answer the multi-choice questions (MCQs).
The questions asked were designed using standard rules for
II. MATERIALS AND METHODS preparing multiple choice question test. The test is designed
This study presents a framework which will use the with the help of a field expert. Students were instructed, to
concepts of educational psychology, neurosciences, be honest as in there is no grading involved. For validation,
technology, statistics, and computational methods to identify the answers were verified with the EEG.
the learning style of a student. We use 2D-based educational
B. Block diagram
tools on learning processes.
A. Experimental Procedure
Subjects: Data is collected from thirty-four healthy
participants in the form of an experiment with written
consent (age range [18-28], male and female, undergraduate
and graduate). Before conducting the experiment, the study
is presented in front of an ethical committee for approval
[13].
Tasks: Subjects were asked to learn and retain contents
which are displayed to them through 2D tools.
Procedure:
1. Before starting the experiment, Participants were
asked to fill the questionnaire, certifying they were Figure 1. Proposed block diagram of the system
healthy (to minimize the effects of non-experimental
variables). The block diagram is defined mathematically as follow.
2. To identify the uniform IQ level of the participants. 𝑃𝑆𝐸𝑖,𝑗 = {𝑃𝑆𝐸𝑖,1 , … , 𝑃𝑆𝐸𝑖,128} (1)
They were asked to perform an intelligence task 𝑃𝑆𝐸𝑖,𝑗 = 𝑃𝑜𝑤𝑒𝑟 𝑠𝑝𝑒𝑐𝑡𝑟𝑢𝑚 𝑒𝑛𝑡𝑟𝑜𝑝𝑦 𝑜𝑓 𝑐ℎ𝑎𝑛𝑛𝑒𝑙𝑠 𝑗 𝑑𝑢𝑟𝑖𝑛𝑔 𝑙𝑒𝑎𝑟𝑛𝑖𝑛𝑔 𝑖
3. At first EEG data was recorded without performing 𝑖 = 𝑙𝑒𝑎𝑟𝑛𝑖𝑛𝑔 1 𝑎𝑛𝑑 𝑙𝑒𝑎𝑟𝑛𝑖𝑛𝑔 2
any task with eyes closed and eyes open condition. 𝑗 = 𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑐ℎ𝑎𝑛𝑛𝑒𝑙𝑠
4. 2D contents were then shown to the participants, 𝐹𝐶𝐿𝑖,𝑀 = {𝐹𝐶𝐿𝑖,1 … 𝐹𝐶𝐿𝑖,𝑀 } (2)
which they must learn. Where EEG data is collected 𝐹𝐶𝐿𝑖,𝑀 = 𝑆𝑒𝑡 𝑜𝑓 𝑠𝑒𝑙𝑒𝑐𝑡𝑒𝑑 𝑓𝑒𝑎𝑡𝑢𝑟𝑒𝑠 𝑑𝑢𝑟𝑖𝑛𝑔 𝑙𝑒𝑎𝑟𝑛𝑖𝑛𝑔 𝑖
during the learning task. 𝑀 = 𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑠𝑒𝑙𝑒𝑐𝑡𝑒𝑑 𝑓𝑒𝑎𝑡𝑢𝑟𝑒𝑠 𝑙𝑒𝑎𝑟𝑛𝑖𝑛𝑔
5. The data is collected in two sessions called as 𝑀𝐿𝑖,𝑙 = {𝑀𝑖,1 … , 𝑀𝑖,𝑙 } (3)
Learning session one (learning 1) and Learning
𝑀𝐿𝑖,𝑙 = 𝑆𝑒𝑡 𝑜𝑓 𝑙𝑒𝑎𝑟𝑛𝑖𝑛𝑔 𝑖 𝑐𝑙𝑎𝑠𝑠𝑖𝑓𝑖𝑒𝑟 𝑝𝑜𝑖𝑛𝑡𝑠
session two (learning 2).
𝑙 = 𝑇𝑜𝑡𝑎𝑙 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑐𝑙𝑎𝑠𝑠𝑖𝑓𝑖𝑒𝑑 𝑝𝑜𝑖𝑛𝑡𝑠
6. The participants had to take test based on learned
contents at the end of session.
The selected format of the test is Multiple choice question
(MCQ) type. MCQ is the best choice for EEG recording. III. RESULTS AND DISCUSSION
Because this type of test has reduced artifact as compared to The behavioral data is analyzed to distinguish visual
other methods such as oral and written test in which other learners and non-visual learners. For learning, the correct
areas of the brain are active because of the sensory responses and reaction times are computed for each
movement which induces artifacts. Following was the participant. The percentage of correct responses per
format of the test. The participants were shown the video of participant was then used to measure his/her learning
8-minutes duration referred to science concepts. The performance. To assess the learning ability, the subjective
contents of animation were: 1) “Human skull and its scores of each participant are calculated. The scores are out
of 100 with a median score of 65 for all participants. Based
*This work was supported by the HiCoE grant for CISIR (Ref. No: 0153CA-002) on the median score, the participants who scored equal or
from the Ministry of Education (MOE), Malaysia. The authors are with the Centre for
Intelligent Signal and Imaging Research (CISIR), Department of Electrical and above the median are considered as visual learners. To
Electronic Engineering, Universiti Teknologi PETRONAS, 32610 Seri Iskandar, Perak.
(* Corresponding author: Ibrahima Faye, Ph.D.; e-mail: [email protected]).
classify the visual learner and non-visual learner, we
analyze the data recorded at two learning sessions called as
Learning session 1 (Learning 1) and Learning session 2
(Learning 2).
To classify good learners and bad learners first Power
Spectrum Entropy(PSE) features are extracted. For feature
selection, the PCA is used to best describe the variance in
the features and to reduce their dimensionality. The
extracted features are of size [34 X 128], and the selected
features is of size [34 X 28].
To evaluate the model performance, the accuracy,
sensitivity (true positive rate), and specificity (true negative
rate) parameters are computed, and the receiver operating
characteristic (ROC) curve is obtained. To calculate the
accuracy, sensitivity, and specificity, the confusion matrix is
first computed. A confusion matrix is used to describe the
performance of a classification model on a set of data with Figure 3. ROC for classification for beta waves(learning1) with AUC of
0.95
known true values.
The ROC is generated by plotting the TP along the y-axis
Mathematically, the accuracy, sensitivity, and specificity
and the FP rate along the x-axis using Matlab. Figure 4
parameters are shown in equations (7)– (9):
𝑇𝑃+𝑇𝑁 shows the ROC curve of the k-NN classifier for the alpha
𝐴𝑐𝑐𝑢𝑟𝑎𝑐𝑦 = ( ) × 100% (7) waves of threshold value (0, 0.5,1) for the alpha waves.
𝑇𝑃+𝑇𝑁+𝐹𝑃+𝐹𝑁
𝑇𝑃 Figure 5 shows the ROC curve of beta waves of threshold
𝑆𝑒𝑛𝑠𝑖𝑡𝑖𝑣𝑖𝑡𝑦 = ( ) × 100% (8)
𝑇𝑃+𝐹𝑁 values (0,0.5,1). The area under the curve (AUC) values for
𝑇𝑁
𝑆𝑝𝑒𝑐𝑖𝑓𝑖𝑐𝑖𝑡𝑦 = ( ) × 100% (9) alpha and beta waves for learning session 1 is 0.97 and 0.94.
𝑇𝑁+𝐹𝑃
The ROC is generated by plotting the TP along the y-axis
and the FP rate along the x-axis using Matlab. Figure 2
shows the ROC curve of the k-NN classifier for the alpha
waves of threshold value (0, 0.5,1) for the alpha waves.
Figure 3 shows the ROC curve of beta waves of threshold
values (0,0.2,1). The area under the curve (AUC) values for
alpha and beta waves for learning session 1 is 0.96 and 0.95.

Figure 4. ROC for classification for alpha waves(learning2) with AUC of

0.97

Figure 2. ROC for classification for alpha waves(learnin1) with AUC of

0.96
 [8] E. O'Rourke, K. Haimovitz, C. Ballweber, C. Dweck, Z. Popovi,
and #263, "Brain points: a growth mindset incentive structure
boosts persistence in an educational game," presented at the
Proceedings of the 32nd annual ACM conference on Human
factors in computing systems, Toronto, Ontario, Canada, 2014.
[9] H. S. Schroder, T. P. Moran, M. B. Donnellan, and J. S. Moser,
"Mindset induction effects on cognitive control: A
neurobehavioral investigation," Biological psychology, vol. 103,
pp. 27-37, 2014.
[10] C. Dweck, "Mindsets and malleable minds: Implications for
giftedness and talent," Malleable minds: Translating insights
from psychology and neuroscience to gifted education, pp. 7-18,
2012.
[11] E. Akioka and L. Gilmore, "An intervention to improve
motivation for homework," Journal of Psychologists and
Counsellors in Schools, vol. 23, pp. 34-48, 2013.
[12] R. M. Ryan and E. L. Deci, Self-determination theory: Basic
psychological needs in motivation, development, and wellness:
Guilford Publications, 2017.
[13] H. U. Amin, A. S. Malik, N. Kamel, W.-T. Chooi, and M.
Hussain, "P300 correlates with learning & memory abilities and
fluid intelligence," Journal of NeuroEngineering and
Figure 5. ROC for classification for beta waves(learning2) with AUC of Rehabilitation, vol. 12, p. 87, September 23 2015.
0.94

IV. CONCLUSION
In this paper we distinguish visual learners from non-visual
learners by analyzing the data collected during learning the
new content. The neural activity of the students is observed
in whole brain. The Power Spectrum entropy(PSE) features
are extracted and feature selection is done using PCA for
training and testing of k-NN classifier. The k-NN classifier
classify visual learners and non-visual learners with 96%
accuracy for alpha and 95% beta bands for learning session
one (L1) and 97% and 94% for learning session two (L2).

REFERENCES

[1] H. M. Truong, "Integrating learning styles and

adaptive e-learning system: Current developments, problems
and opportunities," Computers in Human Behavior, vol. 55, pp.
1185-1193, 2016.
[2] S. Ahn, M. Kim, and D. Ahn, "Relationships between Cognitive
and Learning Styles of Premedical Students," Korean journal of
medical education, vol. 22, pp. 7-13, 2010.
[3] R. E. Mayer and L. J. Massa, "Three facets of visual and verbal
learners: Cognitive ability, cognitive style, and learning
preference," Journal of educational psychology, vol. 95, p. 833,
2003.
[4] A. Abid, I. Kallel, and M. B. Ayed, "Teamwork construction in
E-learning system: A systematic literature review," in 2016 15th
International Conference on Information Technology Based
Higher Education and Training (ITHET), 2016, pp. 1-7.
[5] V. De Pascalis, V. Varriale, and L. D'Antuono, "EEG and
coherence responses to monetary gain and loss during a memory
task: Effects of attentional impulsivity and learning ability,"
International Journal of Psychophysiology, vol. 77, p. 209,
2010/09/01/ 2010.
[6] A. H. a. C. FRASSON, "Predicting Learner Answers
Correctness through Brainwaves Assesment and Emotional
Dimensions " 2009.
[7] H. Rasey, J. F. Lubar, A. McIntyre, A. Zoffuto, and P. L.
Abbott, "EEG Biofeedback for the Enhancement of Attentional
Processing in Normal College Students," Journal of
Neurotherapy, vol. 1, pp. 15-21, 1995/12/01 1995.

View publication stats