Manuscript

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 29

Estimating Cognitive Load During Mental Arithmetic Task

Using EEG Signal

Gunda Manasaa, Krashana.D. Nirdeb, Suhas.S.Gajre c,Ramchandra Manthalkard


a,b,c,d
Department of Electronics & Telecommunication Engineering, SGGSIE &T, Nanded-431606, India.

gundamanasa89@gmail.com, nirde.krishna@gmail.com, suhasgajre@gmail.com, rmanthalkar@gmail.com

Abstract

Cognitive load on the brain is induced by employing cognitive tasks of varying degrees. The load is measured by

using the recorded EEG signal while performing the cognitive task. The EEG signals are processed to determine the

complexity of the task. The current study looks at different Machine learning Classification techniques to classify 2

tasks. Signal from 6 pairs of EEG electrodes in 10-20 system are used for this work. The pairs are located in

different lobes namely pre-frontal, frontal, parietal, temporal, central, occipital. The signal is decomposed using

biorthogonal wavelet basis. Four features like band power, mean, energy and relative energy are computed and

applied as an input to classifier for classification of 2 tasks i.e., resting state and mental arithmetic tasks in individual

brain lobe areas and pair-wise electrode combination. The result shows that, in each individual brain lobe area

frontal and parietal areas have achieved highest classification accuracy of 73.5% while in pair-wise electrode

combination, 3-pair combination (average of all the classifier) i.e., prefrontal, frontal, central have achieved good

classification accuracy of 66%. In EEG band ratios indices, band power as a feature is used to extract different EEG

band ratios for pre-frontal and frontal electrodes in L/R (left/right hemisphere), R/L (right/left hemisphere) and (L-

R)/(L+R) ratios. The energy in different frequency bands is used to quantify the cognitive load. We conclude from

this study that band power ratios of right lobe to left lobe perform better in assessing the cognitive load.

Keywords: Electroencephalogram; Cognitive load; Classification; Prefrontal; Frontal; Parietal; Band power;

Hemisphere; Band ratio indices.


1. Introduction

EEG data is used in a wide range of applications. Acquisition of EEG data requires a good environment to improve

the signal to noise ratio, which helps in analyzing the acquired data. The usual practice is using scalp electrodes to

collect the EEG data. This is a non-invasive method to capture the electrical signals from the brain. Each electrode

collects the data from millions of neurons as a weighted sum of the spatial locations on scalp [1]. Neurons are

activated and deactivated based on the cognitive process involved in the brain. The neural activities generated in the

cognitive process are reflected in captured EEG signal [2]. Generally, brain activity produces the range of waves per

second at different levels, this activity is known as neurological activity. For example, the brain emits low and high

frequencies during sleep and awake conditions [3]. During cognitive tasks, temporal and spectral variations in EEG

signal are crucial for assessing cognitive load.

The human brain responds to the external stimulus in a consistent manner for most people. The processing of signals

is specific for every individual when performing the task with different load (i.e., complex and simple) based on

their skills and aptitude [4]. Cognitive process involves mental effort to perform particular task. With the concept of

Sweller’s [5] “theory of cognitive load focuses on working memory”, whereas Mayer’s [6] explains “theory of

multimedia learning”. “National Information Processing System” is an information processing paradigm, used to

improve cognitive sciences in multimedia environments. Working memory involves carrying out the processing of

information and shows different response patterns when individuals perform the task repeatedly. Participants mental

activity in performing the critical tasks depends on the mental schemes previously stored in long term memory

[6][7].

The electrodes on the scalp are categorized in four major areas depending on the place in the brain from where

signals emanate i.e., frontal, parietal, temporal and occipital. EEG signals are analyzed by various techniques

namely Event related potential (ERP), Event related desynchronization (ERD), Event related synchronization (ERS),

Theta and alpha ratio (TAR), Intermediate alpha frequency (IAF), Individual band powers, Implementing machine

learning and deep learning.

To detect the changes in cognitive load, ERD techniques are analyzed, Klismech found “spectral power in theta

band increases while the spectral power in alpha band decreases” [8]. Recent literature has explored the use of ERP,

ERS and ERD techniques to measure cognitive load. Trammell et. al have estimated the cognitive load and age by
TAR technique which is the ratio of “spectral power of theta band in middle frontal area (Fz) to spectral power of

alpha band in central parietal area (Pz)”. Cognitive load is used to identify the mental cost of performing the task.

This performance can be easily measured with EEG tool which would help in obtaining a good match between the

systems performance and users ability [9][10]. High temporal resolutions and spectral powers of EEG signals are

analyzed at physiological level for recognizing cognitive processes [11][12]. In cortical networks neurons are

activated based on the difficulty of the task performed by the brain [13].

Five frequency bands used to process the EEG signals are: delta (0.5 to 4Hz), theta (4 to 8Hz), alpha (8 to 12Hz),

beta (12 to 30Hz) and gamma (30 to 60Hz). Delta band power increases with an increase in mental workload.

Mostly delta band is observed during the attention states maintained internally by the user while processing the

information during the task. Alpha and theta bands are increased when new information is encoded into the brain.

Higher working memory in frontal midline regions was reported to have increment in theta and decrement in beta

powers. Beta and alpha frequency bands are used to measure the cognitive load activities in fronto-central, frontal

and temporal regions of the cerebellum cortex. Increase in beta and alpha band power shows increase in mental

workload [14][15][16][17][18]. Increase in alpha band power in parietal with electrodes P7 and P8, frontal with

electrodes AF3, AF4, F3 and F4, temporal with electrodes T7 and T8 are used as an indicator to measure mental

workload [8][16]. Similarly same with beta band power, increase in cognitive load in fronto-central with electrodes

Fz and Oz, temporal with electrodes T7 and T8, occipital with electrodes O1 and O2 regions are represented [19].

Temporal region T7 and T8 electrodes are used to measure the delta activity which is more prominent in attention

visual of the task. PSD in each frequency band is analyzed by obtaining the features in frequency domain.

Drowsiness indicated with increase in alpha and theta band [20][21]. Wakefulness and alertness are a sign of

increase in beta band. “θ/α and β/α” ratios are proposed by Eoh et.al. [22]. “(θ + α)/β, θ/β and (θ + α)/ (α + β)” ratios

are proposed by Jap et.al. [23]. “γ/δ and (γ + β)/ (δ + α)” ratios are proposed by da Silveira [24]. All these frequency

band ratios are collected and analyzed from a single brain region.

Working memory and attention abilities are associated with alpha and theta bands. Activation of default mode

network (DMN) over posterior electrodes shows greater synchronization of alpha band [25]. Desynchronization of

alpha band are observed as a natural resource of engagement. Theta frequency band analyzed over frontal cortex

with frequency range 4 to 8 Hz are related to working memory for greater synchronization with arrangement of

neural resources. Better performances in theta frequency bands are observed during a working memory task [26].
Synchronization of theta band is mainly observed in working memory tasks but can also be linked to other functions

like complex non-verbal problem solving and arithmetic strategy [27][28]. Bands over frontal electrodes show

increment in theta and decrement in alpha over posterior and frontal electrode site which would be expected if

targeted computerized executive functioning training results in neurophysiological change.

We mostly use supervised machine learning to classify outputs that are restricted to a limited set of values. Brain

lobe areas classification are not yet studied. Features like band power, mean, energy, relative wavelet energy are

extracted from individual electrodes (12 electrodes) from all brain lobe areas (6 i.e., prefrontal, frontal, central,

parietal, temporal and occipital) when subjects performed two individual tasks (i.e., resting task and mental

arithmetic task). From each brain lobe area, one electrode from left and one electrode from right are picked and

formed a pair combination. Each pair wise combination (ex: F3 and F4) are labelled in a fashion that resting task is

labelled as ‘0’ and mental arithmetic task is labelled as ‘1’ (ex: F3 and F4 of resting task are labelled as ‘0’ and F3

and F4 of mental arithmetic task are labelled as ‘1’) then finally labelled pair wise combinations are applied as input

to classifier for classification.

In the literature, the effect of cognitive load while performing the various arithmetic tasks and the variations in brain

lobes are explored by applying different load levels, as the load increases variations in different brain lobes increases

are studied. Performance enhancement, arousal index, neural activity, engagement, load index, alertness and

cognitive workload index are all EEG band ratio indices used to measure the cognitive load in pre and post

meditation are studied in [43]. Measuring EEG activity indices to see the effect of cognitive load at prefrontal and

frontal brain lobe areas when subjects performed different tasks are not yet studied. So, we consider analyzing the

prefrontal (Fp1, Fp2) and frontal (F3, F4) brain lobe area in L/R(Left/Right), R/L(Right/Left) and L-R/L+R ratios

and assessed EEG band ratio indices to observe the variations in cognitive load when subjects performed two

individual tasks (i.e., resting task and mental arithmetic task).

The main contribution of the proposed research work are:(a) to observe the classification accuracy results for

individual brain lobe areas (i.e., prefrontal, frontal, central, parietal, temporal and occipital).(b)to observe the

classifiers performance when all brain lobe areas(6 pair electrodes) are combined and then excluding one by one

brain lobes(pair electrodes).(c)to measure cognitive load when subjects performing two individual task(i.e., resting

task and mental arithmetic task) can be analyzed by EEG band ratio indices at prefrontal and frontal brain lobe areas
in L/R(Left/Right),R/L(Right/Left) and L-R/L+R ratios. The paper consists of six sections. Section 2 includes

Materials, Section 3 includes Methodology, Section 4 includes Results, Section 5 includes Discussion and Section 6

includes Conclusion.

2. Material

2.1 Recording and Selection of the EEG signals

The EEG signals were recorded from Neurocom Monopolar EEG 23-channel system (Ukraine, XAI-MEDICA). All

23 electrodes are mounted on the head of the subject according to the International 10-20 scheme. The electrodes

mounted on the head record the signals from different parts of the brain, where it is divided into different lobes like

frontal, temporal, parietal, occipital. The electrodes used for measuring frontal area, temporal area, parietal area,

occipital area and central area of the brain are shown below Figure.1. The interconnected ear reference electrodes

are used to reference all electrodes. The electrode impedance was kept lower than 5kΩ, and the sampling rate is

500Hz. A 0.5Hz cut-off frequency HPF, a 45Hz cut-off frequency LPF, and a 50Hz power line notch filter is used.

Fig. 1 10-20 International system of EEG 21 channel Electrode placement.

Recording of each EEG signals includes artifacts-free EEG segments of 180 sec (resting state) and 60 sec (mental

counting). Out of 66 participants 30 were removed from the database due to imprudent artifacts present with eyes

and muscle. So overall we found 36 subjects best. The task includes both Female and Male marked as ‘F’, ‘M’. 27

female and 9 male subjects with age 17-26 with no physical and mental disorders.

2.2 Attribute of Subjects Participation in the protocol

Recording of the EEG signal is shown in below Figure.2 , subjects were asked to sit in a dark insulated chamber

with reclined chair, subjects asked to relax during the resting state or before the task and briefly explained about
arithmetic task, this relaxation state recorded for 3 mins duration. After relaxation, subjects performed the arithmetic

task for 4 min duration but for analyzing the changes occur in the load, the dataset considers 1 min arithmetic task

and remaining 3 mins are excluded. Based on count of subtractions done by the subject’s they were divided into 2

groups: “Bad” and “Good”. A total of 26 subjects performance was under “Good” and 10 subjects performance was

under “Bad”. The subjects participation in the protocol includes series of subtraction or Number of subtractions. It

includes the process of first taking the difference between the four-digit number and the result of subtracting,

divided by the subtrahend. More information can be found in [29].

Fig. 2 EEG data recording during the experiment protocol. Rounding boxes represent the two EEG

recordings stored in the database.

3. Methodology

3.1 Wavelet

Time and frequency domain signals are analyzed simultaneously by Wavelet transform. Wavelets are time-limited

waveforms that exist only for a specific time. Mostly time and frequency domain signals are aperiodic and noisy and

are examined by wavelets. The word “wavelet” means a “small wave”. Based on application requirement wavelets

are categorized into various types. Wavelet transform is a technique where a signal can be decomposed by

considering short duration of interval on which a short wavelet is superimposed. Translating the decomposed signal

to original signal is called inverse wavelet transform. Wavelets are manipulated in two ways:1. translation and 2.

Dilation. Entire signal is converted to wavelets in the next interval of time, which can be known as translation.

Dilation is a method where mother wavelet is enlarged or reduced depending on the frequency of the signal [30].
3.1.1 CWT (Continuous Wavelet Transform)

A wavelet family with mother wavelet 𝜓 (t) consists of functions 𝜓a,b(t) of the form are taken from[31].

1 𝑡−𝑏
𝜓𝑎,𝑏 (𝑡) = 𝜓( ) (1)
√𝑎 𝑎

where a (defines the scale) and b (defines the shift) are positive and real numbers. When |a|>1 then 𝜓a,b(t) has a

longer time width than 𝜓 (t) and relates to decrease in frequency, whereas |a|<1 gives the wavelet obtained from

eq.(1) is high frequency and acquired from mother wavelet[32].

CWT of a function x(t), introduced by Morlet, is defined by


CWT (a, b) = ∫−∞ 𝑥(𝑡) 𝜓𝑎,𝑏 (𝑡)𝑑𝑡 (2)

CWT (a, b) have 2 variable functions:

(1) dilation variable a

(2) time shift variable b [33]

3.1.2 DWT (Discrete Wavelet Transform)

By breaking down the signal into a rough estimation and finer details, the DWT analyses the signal at various

frequency bands with various resolutions. Scaling functions and wavelet functions, which are related to low pass and

high pass filters, respectively, these are two sets of functions used by DWT. The time domain signal is simply

subjected to consecutive lowpass and high pass filtering to obtain the signal's decomposition into various frequency

bands. First, a half band HPF (g[n]) and a LPF (h[n]) are applied to the input original signal x[n]. Therefore, the

signal can be subsampled by two by simply removing every other sample. This is a single decomposition level and

may be stated mathematically as follows:

Ylow[k] =∑𝑛 𝑥[𝑛] . h[2k-n] (3)

Yhigh[k] =∑𝑛 𝑥[𝑛] . g[2k-n] (4)

where Yhigh[k] highpass filter output and Ylow[k] lowpass filter output.

Below Figure.3 shows Discrete Wavelet Decomposition, where x[n] original input signal to be decomposed, h[n]

LPF and g[n] HPF.Biorthogonal and orthogonal are compactly supported by DWT and characterized by the high and

low pass filters for synthesis. Few wavelet families are available they are: Meyer, Biorthogonal, Daubechies, Haar,
Coiflets, Symlets, Morlet and Mexicanhat[34]. Morlet, Meyer and Mexicanhat are symmetric wavelets in shape,

while Symelts, Coiflets, Haar and Daubechies are orthogonal supported wavelet.

Fig. 3 Discrete Wavelet Decomposition.


3.1.3 Biorthogonal Wavelets:

The family of orthogonal wavelets is expanded by biorthogonal wavelets. Biorthogonal wavelets with periodic

nature of signals are analyzed by products of vector matrices and structured matrices [35][36]. FIR filters are used

for reconstruction and decomposition when signal is regular. To avoid this difficulty, wavelet with dual scaling will

have the following properties:

1. Calculation algorithms maintained are simple and Outside of a segment, they are zero.

2. The associated filters are symmetrical.

3. Calculation functions are simpler to create than those used in Daubechies wavelets.

For reconstruction and decomposition two wavelets are used, biorthogonal wavelets exhibit the property for signal

and image reconstruction of linear phase. Artifacts at the borders of wavelet sub bands are caused by asymmetric

nature of Biorthogonal wavelets [37].

We choose the biorthogonal 4.4 wavelet for this study. The orthogonal wavelet transform has the advantage of being

orthogonal because the associated wavelet transform is orthogonal, and thus the inverse wavelet transform is the
adjoint of the wavelet transform. When compared with orthogonal wavelet more degree of freedom is allowed by

BWT (biorthogonal wavelet transform) [38]. Functions and filter for decomposition are shown in Figure 4. and

Figure. 5 below, respectively.

There are two scaling functions in biorthogonal they are: 𝜙 ,𝜙̃ which may result in two distinct multiresolution

analyses and as a result, two distinct wavelet functions 𝜓, 𝜓̃. As a result, the coefficient numbers M and N in the

scaling sequences 𝑎, 𝑎̃ may differ. Biorthogonality condition must satisfy the scaling sequence [39].

∑𝑛𝜖𝑍 𝑎𝑛 𝑎𝑛+2𝑚 = 2 . 𝛿𝑚,0 (5)

Determination of wavelet sequences is as follows:

𝑏𝑛 = (-1)n 𝑎̃𝑀−1−𝑛 (n=0,1,……..N-1) (6)

𝑏̃𝑛 = (-1)n 𝑎𝑀−1−𝑛 (n=0,1,……..N-1) (7)

Fig. 4 Bior4.4 Filter : (a) LPF ; (b)HPF. Fig. 5. Bior4.4 functions: (a) scaling function;
(b) wavelet function

3.2 Feature Extraction

The prominent frequency components from EEG which are important for the analysis, named as delta, theta, alpha,

beta, and gamma are found and the features are extracted.
We have considered two tasks; the first task involves resting state (i.e., eyes closing) and the other task involves

performing mental arithmetic task (i.e., series of subtraction task). Statistical parameters like band power, energy,

mean and relative wavelet energy are calculated for individual electrodes like Fp1, Fp2, F3, F4, P3, P4, C3, C4, T3,

T4, O1, O2.

3.2.1 Band Power

Band Power consider the energy present in frequency bands and is defined as a sum of squares of data points [40]

BP = ∑𝑁
𝑛=1 |𝑥𝑛 |² (8)

Where xn are signal data points from 1 to N.

3.2.2 Energy

If the total energy E is finite, then a signal is said to be an energy signal

The energy of a discrete-time signal x[n] is

𝐸𝑥 = ∑∞
𝑛=−∞ |𝑥[𝑛]|
2
(9)

3.2.3 Mean

Mean is defined as ratio of sum of elements by total number of elements:

1
Mean = ∑𝑛𝑖=1 𝑎𝑖 (10)
𝑛

where n = number of values, 𝑎𝑖 = data set values

3.2.4 Relative Wavelet Energy

Relative wavelet gives information regarding the energy present in the frequency bands which is used to detect the

similarities between segments of an EEG signal. In this we have determined the RWE(Relative Wavelet Energy) for

all in a resting state task and mental arithmetic task. Usually, we considered 6 pairs of electrodes from 6 lobe areas

and determined RWE. The energy at each decomposition level is given by:

𝐸 j = ∑𝑘 |𝑑𝑗,𝑘 |2 j = 1,…………,N (11)

𝐸 N+1 = ∑𝑘 |𝐶𝑘 |2 (12)


after wavelet decomposition, total energy of the signal is obtained by

𝐸 total = ∑𝑁+1
𝑗=1 𝐸 j (13)

RWE is defined as,

𝐸𝑗
𝜌j = j = 1,….,N+1 (14)
𝐸𝑡𝑜𝑡𝑎𝑙

Fig.6 Basic Block diagram for classification of resting state and mental arithmetic task.

4. Results

In this section, extracted 4 features are applied in two methods (Individual Brain Lobe Areas Classification

and Pair Wise Electrode Combination Classification output) and 1 feature in one method (EEG Band Ratios).
4.1 Individual Brain Lobe Areas Classification

We consider pair of electrodes from each lobe. Electrodes placed on the scalp are arranged in a manner

such that it collects the data from 4 main lobes of the brain. For our convenience we divided the 4 brain lobe areas

into 6 brain lobe areas (i.e., prefrontal, frontal, central, temporal, parietal and occipital). From each area we consider

one channel from the left hemisphere and another from the right hemisphere. So, finally we obtained 6 pairs of

electrodes from 6 areas of the brain. We have four features i.e., band power, energy, mean and relative energy are

applied to the classifier with standard scalar for classification. Machine learning classification algorithms use input

training data to expect the output data to fall in which category. We implemented different classifiers like Support

Vector Machine (SVM), K-Nearest Neighbor (KNN), Random Forest (RF), Logistic Regression (LR), Naive Bayes

(NB), Decision Tree (DT) on each brain lobe areas for classification of two task (i.e., resting state task and mental

arithmetic task). Resting task labelled as ‘0’ and Mental Arithmetic task is labelled as ‘1’. Individual brain lobe

areas performance in classification can be seen from accuracy table and percentage graphs shown below.

From below Figure.7, Fp1& Fp2 electrodes have produced high accuracy in Random Forest with 82% and low

accuracy in Logistic Regression with 62%. While precision and recall show the same results (i.e., Random Forest).

(a) (b)

Fig.7 Classification of Prefrontal brain lobe area (Fp1& Fp2 electrode pair) (a)Accuracy Table, (b)Percentage

Graph.

From below Figure.8, F3& F4 electrodes have produced high accuracy in Random Forest with 86% and low

accuracy in Decision Tree with 65%. While precision has predicted the output with specific category in Decision

Tree and Random Forests, recall represents the how many times the model able to detect a certain category can be

seen high in Logistic Regression and low in Decision Tree.


(c) (d)

Fig.8 Classification of Frontal brain lobe area (F3 & F4 electrode pair) (c)Accuracy Table, (d)Percentage

Graph.

From below Figure.9, C3 & C4 electrodes have produced high accuracy in Random Forest with 82% and low

accuracy in Naive Bayes with 41%. While precision predicted highest in Naive Bayes and lowest in Decision Tree

and KNN, whereas in recall highest in Random Forests and lowest in Naive Bayes.

(e) (f)

Fig.9 Classification of Central brain lobe area (C3 & C4 electrode pair) (e)Accuracy Table, (f)Percentage

Graph.

From below Figure.10, P3 & P4 electrodes have produced high accuracy in Random Forest with 82% and low

accuracy in Support Vector Machine with 58%. While precision predicted highest in Random Forests and Naive

Bayes and lowest in Support Vector Machine whereas in recall highest in Random Forests and lowest in Support

Vector Machine.
(g) (h)

Fig.10 Classification of Parietal brain lobe area (P3 & P4 electrode pair) (g)Accuracy Table, (h)Percentage

Graph.

From below Figure.11, T3 & T4 electrodes have produced high accuracy in Logistics Regression with 86% and low

accuracy in Naive Bayes with 44%. While precision predicted highest in Naive Bayes and lowest in Decision Tree

and KNN, whereas in recall highest in Logistics Regression and lowest in Naive Bayes.

(i) (j)

Fig.11 Classification of Temporal brain lobe area (T3 & T4 electrode pair) (i)Accuracy Table, (j)Percentage

Graph.

From below Figure.12, O1 & O2 electrodes have produced high accuracy in Random Forests with 83% and low

accuracy in Decision Tree with 55%. While precision predicted highest in Naive Bayes and lowest in Logistics

Regression, Support Vector Machine and Decision Tree, whereas in recall highest in Random Forests and lowest in

Naive Bayes.
(k) (l)

Fig.12 Classification of Occipital brain lobe area (O1 & O2 electrode pair) (k)Accuracy Table, (l)Percentage

Graph.

From all brain lobe areas, frontal (F3 & F4 pair) and parietal (P3 & P4 pair) obtained the high classification

accuracy with 73.5% and low classification accuracy with 61.6% seen in central lobe area (C3 & C4 pair).

4.2 Pair Wise Electrode Combination Classification output

Extracted features are utilized for obtaining classification for 6 pair wise electrode combination and excluding one

by one pair to observe the changes in classification output with different classifiers. Below Table 1 shows the

representation of different classifiers performance in pair wise combinations.1st column represents all 6 pair

combinations of electrodes (i.e., prefrontal (Pf), frontal(F), central(C), temporal(T), parietal(P), occipital(O)), these 6

pair combinations are applied as a input to different classifier and observed classification output. Highest accuracy

found in Random Forest with 72% and lowest accuracy seen in Naive Bayes with 50%. 2nd column we consider 5

pair combinations (i.e., prefrontal (Pf), frontal(F), central(C), temporal(T), parietal(P)) and repeated same process

(for all pair combinations) by applying the input and observing the output of different classifier. High accuracy was

found in Random Forest with 75% and lowest accuracy in Naive Bayes with 43%. 3 rd column we consider 4 pair

combinations (i.e., prefrontal (Pf), frontal(F), central(C), temporal(T)) and observed high accuracy in Random

Forest with 76% and lowest accuracy in Naive Bayes with 45%. 4 th column we consider 3 pair combinations (i.e.,

prefrontal (Pf), frontal(F), central(C)) and observed high accuracy in Decision Tree with 73% and lowest accuracy

in SVM with 56%. Finally, in the 5th column we consider 2 pair combinations (i.e., prefrontal (Pf), frontal(F)) and

observed high accuracy in Random Forest with 72% and lowest accuracy in SVM with 53%.
4 pair 3 pair 2 pair

All 6 pair electrodes 5 pair electrodes electrodes electrodes electrodes

combination combination combination combination combination

Classifier (Pf+F+C+T+P+O) (Pf+F+C+T+P) (Pf+F+C+T) (Pf+F+C) (Pf+F)

Logistic

Regression 58% 56% 67% 65% 62%

SVM 56% 57% 68% 56% 53%

Decision Tree 57% 64% 56% 73% 68%

Random Forest 72% 75% 76% 70% 72%

Naive Bayes 50% 43% 45% 70% 58%

KNN 68% 63% 74% 62% 70%

Table 1 Representation of different classifiers performances in pair wise electrode combinations.

We observe that from above Table 1, we averaged each pair combinations with different classifiers and found that 3

pair combination (i.e., prefrontal, frontal, central) has obtained 66%, while least accuracy with 59.6% observed in 5

pair combination. We also observed that from all classifiers Random Forests classifier has performed better and

Naive Bayes classifier performed worst.

4.3 EEG Band Ratios

EEG band ratios are assessed by using band power which is used to represents the different cognitive indices [41].

The theta, alpha and beta band can be associated with attention, focused executive processing and working memory

retention [42].

Analysis done based on considering the prefrontal and frontal electrodes. Prefrontal electrodes play a vital role in

solving problems, cognitive control functions and reasoning. Frontal electrode functions are memory formation,

attention, planning and self-control. Fp1, Fp2 are prefrontal electrodes and F3, F4 are frontal electrodes considered

for obtaining EEG band ratios. Different EEG band ratios are analyzed to see the performances of prefrontal and

frontal electrodes in resting task and mental arithmetic task. Calculating different EEG band ratios for pre-frontal

and frontal electrodes in L/R, R/L and (L-R)/(L+R) ratios. We summed the band power of Fp1 and Fp2 and
calculated EEG band ratios, then after we found difference for Fp1 and Fp2 with same band power and calculated

EEG band ratios. The sum and difference results are taken ratios and obtained results are shown in below Table 2.

Left (Fp1, F3) and right (Fp2, F4) hemisphere of prefrontal and frontal electrodes and represented in ratio form: (L-

R)/(L+R), L/R, R/L for analyzing the cognitive load.

Performance Enhancement: the ratio of alpha by theta is known as performance enhancement. Subjects

enrichment in performing the task is evaluated by this ratio. When alpha increases and theta decreases, indicates the

state of relaxation and awareness about the task. If alpha decreases and theta increases, indicates subjects are quite

focused and concentrated about the task.

Arousal Index: the ratio of beta by alpha is known as arousal index. Subjects tiredness and variations in tolerance

of sleep disorders in the task are evaluated by this ratio. When alpha increases and beta decreases, subjects show

relaxed state of mind and awareness about the task. If alpha decreases and beta increases, indicates the state of

tiredness, consciousness and concentration on the task.

Neural Activity: it is defined as ratio of beta by theta. Subjects state of conscious and sub-conscious are evaluated

by this activity. Subjects brain activity producing the range of waves per second at different levels is known as

neural activity. When beta increases and theta decrease, indicates anxiety and obsessive in performing the task. If

theta increases and beta decreases, indicates creative and intuitive thoughts in processing the task.

Engagement: it is defined as ratio of beta by alpha plus theta. Subjects involvement in performing the task can be

assessed by this ratio. Increase in beta and decrease in alpha indicate awareness states or slower frequencies. If beta

decreases and theta band increases indicates drowsiness and low attention.

Load Index: it is defined as ratio of theta by alpha. When difficulty levels of the task increase, the load on subjects

brain increases. Increase in alpha and decrease in theta indicates the states of relaxation and full awareness of the

task. If increase in theta and decrease in alpha indicates the states of attention and alertness.

Alertness: it is defined as a ratio of alpha plus theta by beta. Subjects paying attention and ready to respond to

stimulus can be evaluated by alertness. An increase in theta and decrease in beta indicates the state of attention and

alertness. If increase in beta and decrease in theta indicates the states of anxious, stress (if levels increases) and

concentration.
Cognitive Workload Index (CWI): it is the ratio of beta plus gamma bands to alpha plus theta bands. Cognitive

load variations can be easily assessed with this parameter. Increase in beta and gamma shows the presence of

cognitive load based on the level of task. If the level of task increases cognitive load increases. Beta and gamma

bands are represented for higher concentration, stress, tiredness.

Comparison of Cognitive Indices in Resting Task and Mental Arithmetic Task using mean ± Std.
The increase of frontal theta band power shows meditative state and concentration. Increase in alpha and beta

indicates increase in performance enhancement, arousal index, neural activity and engagement.

EEG Band Ratios of Fp1& Fp2 electrode in (L-R)/(L+R) ratio for resting task and mental arithmetic task:

𝛽
From Table 2. In arousal index ⁄𝛼, decrease in beta and increase in alpha shows subjects felt difficult in focusing

and too relaxed in resting task, whereas it showed vice versa in mental arithmetic task in which subjects were active

𝛽
and anxious in performing the task. In neural activity ⁄𝜃, decrease in beta and increase in theta shows subjects

showed meditative states in resting task, whereas increment in 𝛽 states that subjects were alert and concentrated in

𝛽
mental arithmetic task. In engagement ⁄(𝛼 + 𝜃) , decrease in beta and increase in theta plus alpha indicates that

subjects were not focused and alert in resting task, whereas increment in 𝛽 states that subjects were focused and

(𝛼 + 𝜃)
attentional in performing the mental arithmetic task. In alertness ⁄𝛽 , increase in alpha plus theta and

decrease in beta seen in resting task which indicates subjects were in lite awake, relaxed, whereas increment in 𝛽

states that subjects were concentrated and anxious in performing the mental arithmetic task. In cognitive workload

(𝛽 + 𝛾)
index ⁄(𝛼 + 𝜃), decrease in beta plus gamma band and increase in theta plus alpha band indicates subjects

were awake, relaxed and less concentrated in resting task, whereas increase in beta plus gamma band indicates

subjects were focused, alert and concentrated in performing the mental arithmetic task. In performance enhancement
𝛼⁄ , increase in theta and decrease in alpha indicates meditative states of the subjects in resting task, whereas
𝜃

increase in alpha and decrease in theta shows subjects are awake in performing the mental arithmetic task. In load

index 𝜃⁄𝛼, increase in theta and decrease in alpha indicates subjects were focused and concentrated in mental

arithmetic task, while decrease in theta indicates subjects were awake and showed meditative states in resting task.

(Fp1-Fp2)/(Fp1+Fp2) Activity Index Resting Task Mental Arithmetic Task


electrode EEG Band

Ratio (Hz)

𝛼⁄ Performance Enhancement 1.029±0.447 1.666±0.467


𝜃

𝛽⁄ Arousal Index 0.116±0.442 0.918±2.211


𝛼

𝛽⁄ Neural Activity 0.148±0.950 1.561±1.033


𝜃

𝛽 0.177±1.172 1.284±1.275
⁄(𝛼 + 𝜃) Engagement

𝜃⁄ Load Index 0.670±2.233 0.900±2.139


𝛼

(𝛼 + 𝜃) 12.512±0.852 7.849±0.783
⁄𝛽 Alertness

(𝛽 + 𝛾) 0.051±1.177 1.264±1.265
CWI
(𝛼 + 𝜃)

Table 2. Representation of different EEG Band Ratios performances of Fp1& Fp2 electrode in (L-R)/(L+R)

for resting task and mental arithmetic task.

EEG Band Ratios of F3& F4 electrode in (L-R)/(L+R) ratio for resting task and mental arithmetic task:

From Table 3. in performance enhancement 𝛼⁄𝜃, increase in theta and decrease in alpha indicates subjects f showed

meditative states in resting task, whereas increase in alpha and decrease in theta indicates subjects were awake and

𝛽
relaxed in performing the mental arithmetic task. In arousal index ⁄𝛼, increase in beta and decrease in alpha

indicates subjects found difficulty in concentration. Beta showed decrement in resting task and increment in mental

arithmetic task. Subjects felt anxiety and obsessiveness in mental arithmetic task, whereas decrement in resting task

𝛽
indicates subjects were normal and anxious. In neural activity ⁄𝜃 increase in beta and decrease in theta indicates

active thinking and paying attention towards the task in mental arithmetic task whereas it observed to be decrease in

𝛽
resting task. Engagement ⁄(𝛼 + 𝜃) of the subjects decreased in resting task which indicates reduce in alert and

focus, whereas it showed increased in mental arithmetic task which indicates that subjects were concentrated and

found difficulty in performing the task. In load index 𝜃⁄𝛼, decrease in theta and increase in alpha seen in resting task

which indicates that subjects were found to be meditative, whereas it showed increment in mental arithmetic task
which indicates that subjects were awake, observed to have load on brain in performing the task. Alertness

(𝛼 + 𝜃)
⁄𝛽 found to decrease in resting task when compared with mental arithmetic task, beta increased in mental

arithmetic task where subjects found difficulty in problem-solving and are alert in performing the task, whereas it

decreased in resting task which states subjects were normal awake and less concentrated. In CWI

(𝛽 + 𝛾)
⁄(𝛼 + 𝜃)increase in beta plus gamma band and decrease in alpha plus theta band indicates subjects were

highly concentrated, visually attentional and alertness in performing the mental arithmetic, whereas it decreased in

resting task indicates that subjects were less concentrated and attention.

(F3-F4)/(F3+F4) Activity Index Resting Task Mental Arithmetic Task

electrode EEG

Band Ratio (Hz)

𝛼⁄ Performance Enhancement 0.340±0.449 0.981±0.682


𝜃

𝛽⁄ Arousal Index 0.042±2.007 0.621±1.686


𝛼

𝛽⁄ Neural Activity 0.937±0.901 1.610±1.150


𝜃

𝛽 0.159±1.061 0.613±1.291
⁄(𝛼 + 𝜃) Engagement

𝜃⁄ Load Index 1.246±2.226 3.019±1.466


𝛼

(𝛼 + 𝜃) 0.862±0.942 1.629±0.774
⁄𝛽 Alertness

(𝛽 + 𝛾) 1.180±1.064 1.610±1.253
CWI
(𝛼 + 𝜃)

Table 3. Representation of different EEG Band Ratios performances of F3 & F4 electrode in (L-R)/(L+R) for

resting task and mental arithmetic task.

We have analyzed the cognitive load by (L-R)/(L+R) ratio for prefrontal and frontal electrodes of resting task and

mental arithmetic task. The implementation has been extended by considering L/R (left/right hemisphere) and R/L

(right/left hemisphere) ratio for prefrontal and frontal electrodes and executed EEG band ratio output to assess
cognitive load. Below Table 4, Table 5 represents the EEG Band Ratios performances in L/R and R/L ratios of Fp1,

Fp2, F3, F4 electrodes in resting task and mental arithmetic task.

EEG Band Ratios in L/R, R/L ratio of Fp1, Fp2, F3, F4 electrodes for resting task and mental arithmetic

task:

Performance Enhancement 𝛼⁄𝜃: from Table 4. in L/R ratio for Fp1 and Fp2 electrodes, decrease in alpha and

increase in theta indicates subjects were awake and awareness about performing the mental arithmetic task, whereas

it showed vice versa in resting task indicates that subjects felt sleepy. Increase in theta and decrease in alpha in L/R

ratio for Fp1 and Fp2 electrodes were also observed same in L/R ratio of F3 and F4 electrodes from resting task to

mental arithmetic task indicates that subjects performance enhanced from resting task to mental arithmetic task.

From Table 5. R/L ratio for Fp1 and Fp2 electrodes, decrease in alpha and increase in theta observed to have

improved from resting task to mental arithmetic task. The same decrease in alpha and increase in theta observed in

R/L ratio for F3 and F4 electrodes from resting task to mental arithmetic task which indicates subjects were awake,

awareness and attention towards performing the task.

𝛽
Arousal Index ⁄𝛼: from Table 4. L/R ratio for Fp1 and Fp2 electrodes, increase in beta and decrease in alpha

indicates subjects were concentrated and awaked in mental arithmetic task, whereas decrease in beta and increase in

alpha showed that subjects are less awake and relaxed in resting task. The same increase in beta and decrease in

alpha observed in L/R ratio for F3 and F4 electrodes, subjects were concentrated, being in alert and problem-solve

involvement showed increment from resting task to mental arithmetic task. From Table 5. R/L ratio for Fp1 and Fp2

electrodes, increase in beta and decrease in alpha observed to have improved from resting task to mental arithmetic

task which indicates subjects felt anxious, concentrated while performing the mental arithmetic task, whereas

subjects in resting task did not observe the change. The same increase in beta and decrease in alpha observed in R/L

ratio for F3 and F4 electrodes from resting task to mental arithmetic task indicates subjects observed to be anxious,

stressed and felt tiredness in performing the mental arithmetic task.

𝛽
Neural Activity ⁄𝜃: from Table 4. L/R ratio for Fp1 and Fp2 electrodes, increase in beta and decrease in theta

indicates subjects were concentrated, alert and focused on mental arithmetic task, decrease in beta found in resting

task when compared to mental arithmetic task. The same increase in beta and decrease in theta observed in L/R ratio

for F3 and F4 electrodes, subjects were concentrated and attention in mental arithmetic task, while decrease in beta
and increase in theta indicates that subjects felt focused and meditative in performing resting task. From Table 5.

R/L ratio for Fp1 and Fp2 electrodes, increase in beta and decrease in theta observed to have improved from resting

task to mental arithmetic task which indicates subjects felt anxious, concentrated while performing the mental

arithmetic task, whereas subjects in resting task did not concentrate. The same increase in beta and decrease in theta

observed for R/L ratio for F3 and F4 electrodes in resting task to mental arithmetic task, which indicates subjects

were highly concentrated and anxious in performing the mental arithmetic task, whereas subjects did not feel to have

concentrate in performing resting task.

𝛽
Engagement ⁄(𝛼 + 𝜃): from Table 4. L/R ratio for Fp1 and Fp2 electrodes, increase in beta and decrease in theta

plus alpha indicates subjects were concentrated, less sleepy, low in relax in mental arithmetic task, decrease in beta

found in resting task. The L/R ratio of F3 and F4 electrodes, increase in beta and decrease in theta plus alpha states

that subjects concentration has shown the improvement in performing the task from resting task to mental arithmetic

task. From Table 5. R/L ratio for Fp1 and Fp2 electrodes and R/L ratio for F3 and F4 electrodes, have same increase

in beta and decrease in theta plus alpha can be seen in resting task to mental arithmetic task.

Load Index 𝜃⁄𝛼: from Table 4. L/R ratio for Fp1 and Fp2 electrodes, increase in theta and decrease in alpha

indicates subjects were quite focused in performing mental arithmetic task, while it decreases in theta and increase

in alpha indicates subjects were relax and did not find any difficulty in performing resting task. L/R ratio of F3 and

F4 electrodes increase in theta and decrease in alpha observed to have improved from resting task to mental

arithmetic task which indicates subjects are attention and quite focused in performing mental arithmetic task when

compared to resting task. From Table 5. R/L ratio for Fp1 and Fp2 electrodes, increase in theta and decrease in

alpha indicates subjects were quite focused and alertness in performing mental arithmetic task, while it decreases in

theta and increase in alpha indicates subjects were relax and did not find any difficulty in performing resting task.

The same conditions of R/L ratio for Fp1 and Fp2 electrodes are observed in R/L ratio of F3 and F4 electrodes in

resting task and mental arithmetic task.

(𝛼 + 𝜃)
Alertness ⁄𝛽 : from Table 4. L/R ratio for Fp1 and Fp2 electrodes, decrease in beta and increase in alpha

plus theta which indicates subjects were less anxious, less concentrated, less focused and awake in resting task,

while increase in beta and decrease in alpha plus theta indicates subjects were concentrated and anxious in
performing mental arithmetic task. L/R ratio of F3 and F4 electrodes, decrease in beta and increase in alpha plus

theta can be seen in resting task which indicates subjects were relaxed and quite relaxed in resting task, whereas

decrease in alpha plus theta and increase in beta indicates subjects were more concentrated and found difficulty in

performing mental arithmetic task. From Table 5. R/L ratio for Fp1 and Fp2 electrodes, increase in alpha plus theta

can be seen in resting task which indicates subjects were relaxed and attentional, decrease in alpha plus theta and

increase in beta can be seen in mental arithmetic task which indicates subjects were more concentrated, alert and

stressed. R/L ratio of F3 and F4 electrodes, increase in alpha plus theta and decrease in beta seen in resting task

indicates subjects were relaxed and attentional in performing the task, increase in beta and decrease in alpha plus

theta in mental arithmetic task indicates subjects were concentrated, alert and consciousness.

(𝛽 + 𝛾)
Cognitive Workload Index (CWI) ⁄(𝛼 + 𝜃): from Table 4. L/R ratio for Fp1 and Fp2 electrodes,

increment from resting task to mental arithmetic task indicates that increase in beta plus gamma and decrease in

alpha plus theta indicates that subjects were high concentrated, consciousness, visual stimulated, less focused and

relaxed. The same situation of L/R ratio of Fp1 and Fp2 electrodes are observed in L/R ratio of F3 and F4 electrodes

in resting task and mental arithmetic task. From Table 5. The same increase in beta plus gamma and decrease in

alpha plus theta found in R/L ratio for Fp1 and Fp2 electrodes of resting task, the same can be observed in R/L ratio

of F3 and F4 electrodes. The assessment of cognitive load is measured from prefrontal and frontal electrodes. Higher

level cognitive functioning is associated with prefrontal electrodes. The frontal lobe is responsible for executive

functions like problem-solving, reasoning, planning, attention, controlling our behavior, decision-making, emotions,

consciousness, memory. Usually, cognitive load is observed when a critical task is given to a subject. Series of

subtraction is the activity involved to assess the subjects cognitive load in mental arithmetic task, eye’s closing

activity performed by subjects in resting task.

In our study, cognitive load has shown improvement from resting task to mental arithmetic task in (L-R)/(L+R),

L/R and R/L ratio. The ratios L/R and R/L have shown the variations from resting task to mental arithmetic task. We

found that R/L ratio has shown increase in cognitive load when compared with other ratios, as L/R ratio has also

shown increment but not as much as R/L. The ratio (L-R)/(L+R) increased from rest to mental arithmetic task, but

not as much as L/R and R/L. We found that when difficulty in task increases the cognitive load increases and this

has been assessed by L/R, R/L and (L-R)/(L+R) ratios with different EEG band ratios.
EEG Band Activity Index L/R ratio for Fp1 and Fp2 electrodes L/R ratio for F3 and F4 electrodes

Ratio (Hz) Resting Task Mental Arithmetic Resting Task Mental Arithmetic

Task Task

𝛼⁄ Performance 0.972±0.691 0.976±0.830 1.033±0.691 1.127±0.483


𝜃
Enhancement

𝛽⁄ Arousal Index 1.051±2.459 1.065±1.814 0.971±1.230 1.073±1.406


𝛼

𝛽⁄ Neural 1.041±1.700 1.140±1.507 1.004±0.850 1.095±0.680


𝜃
Activity

𝛽 0.528±1.005 0.529±0.823 0.493±0.502 0.516±0.458


⁄(𝛼 + 𝜃) Engagement

𝜃⁄ Load Index 1.028±1.446 1.124±1.203 0.967±1.445 1.078±2.069


𝛼

(𝛼 + 𝜃) 1.893±0.994 1.769±1.214 2.024±1.988 1.935±2.181


⁄𝛽 Alertness

(𝛽 + 𝛾) 1.056±2.189 1.073±1.747 0.989±1.165 1.035±0.840


CWI
(𝛼 + 𝜃)

Table 4. Representation of different EEG Band Ratios performances in L/R ratio of Fp1, Fp2, F3, F4

electrodes for resting task and mental arithmetic task.

EEG Band Activity Index R/L ratio for Fp1 and Fp2 electrodes R/L ratio for F3 and F4 electrodes
Ratio (Hz)
Resting Task Mental Arithmetic Resting Task Mental Arithmetic

Task Task

𝛼⁄ Performance 0.945±0.446 1.007±0.941 0.866±0.218 1.011±0.679


𝜃
Enhancement

𝛽⁄ Arousal Index 1.070±1.981 1.081±1.968 1.050±1.557 1.167±1.299


𝛼

𝛽⁄ Neural 1.018±0.884 1.089±1.852 0.909±0.340 0.978±0.882


𝜃
Activity
𝛽 0.523±0.611 0.542±0.954 0.485±0.279 0.486±0.525
⁄(𝛼 + 𝜃) Engagement

𝜃⁄ Load Index 1.057±2.240 1.692±1.062 1.154±4.578 1.188±1.472


𝛼

(𝛼 + 𝜃) 1.910±1.635 1.842±1.048 2.051±3.582 2.015±1.902


⁄𝛽 Alertness

(𝛽 + 𝛾) 1.093±1.465 1.108±2.195 0.785±0.651 0.959±0.945


CWI
(𝛼 + 𝜃)

Table 5. Representation of different EEG Band Ratios performances in R/L ratio of Fp1, Fp2, F3, F4

electrodes for resting task and mental arithmetic task.

5. Discussion

It is known that bior4.4 decomposition technique used to extract the EEG frequency bands. By analyzing these

bands, we can predict the subjects performance by considering individual electrodes from left and right hemisphere

and obtained 6 pair of electrodes from 6 lobe areas. Here, we considered Fp1-Fp2 as pre-frontal electrodes, F3-F4 as

frontal electrodes, C3-C4 as central electrodes, P3-P4 as parietal electrodes, T3-T4 as temporal electrodes and O1-

O2 as occipital electrodes. The electrodes with odd numbering of electrodes present towards left hemisphere

whereas even numbering of electrodes present towards right hemisphere. In the first method, decomposed signal

used to extract four features from 6 pairs of electrodes and applied to different classifier for classifying resting state

and mental arithmetic state. Prefrontal area of left and right hemisphere are measured with Fp1 and Fp2 and their

features are applied to different classifier, highest accuracy obtained in Random Forests with 82% and lowest

obtained in Logistic Regression with 62%.In frontal area, we obtained highest accuracy obtained in Random Forests

with 86% and lowest obtained in Decision Tree with 65%.In central area, highest accuracy obtained in Random

Forests with 82% and lowest obtained in Naive Bayes with 41%.In parietal area, highest accuracy obtained in

Random Forests with 82% and lowest obtained in Support Vector Machine with 58%.In temporal area, highest

accuracy obtained in Logistic Regression with 86% and lowest obtained in Naive Bayes with 44%. In occipital area,

highest accuracy obtained in Random Forests with 83% and lowest obtained in Decision Tree with 55%. In all brain

lobe areas frontal (F3 & F4 pair) and parietal (P3 & P4 pair) lobe areas have obtained the high classification

accuracy with 73.5% and low classification accuracy seen in central (C3 & C4 pair) with 61.6%. In the second

method, from Table 1. we combined all features of 6 pair of electrodes from 6 brain area and applied as an input to
different classifier for classification, from 6 pair combination we excluded the occipital pair electrodes and again

features are applied as an input to classifier for classification, this process repeated by excluding parietal, temporal

and central electrodes and remained with only 2 pair electrodes i.e., prefrontal and frontal electrodes. Prefrontal and

Frontal lobe areas are used to assess the cognitive control functions. 3 pair combination has obtained better accuracy

results with 66% while least accuracy 59.6% in 5 pair combinations. In all pair wise combinations, Random Forest

classifier has performed good and obtained better accuracy results while Naive Bayes classifier has performed bad

and obtained least accuracy results. In the third method, EEG band ratios are analyzed by considering the left and

right hemisphere of prefrontal and frontal electrodes. The ratios L/R, R/L and (L-R)/(L+R) are implemented to

observe the variations in the bands ratios which are used to assess different functionalities changes occur in the left

and right hemisphere. This is one of the methods to assess the cognitive load measurement when subjects perform

resting task and mental arithmetic task. In Table 2, Table 3 shows EEG Band Ratios performances of prefrontal and

frontal electrode in (L-R)/(L+R) for resting task and mental arithmetic task. Beta band showed decrease in resting

task whereas it found to be increase in mental arithmetic task which can be seen from Table 2, Table 3. When

considering the CWI (Cognitive Workload Index) of Fp1 and Fp2 electrodes, subjects felt difficulty in performing

the mental arithmetic task which is observed by increase in beta and gamma band. In F3 and F4 the same consistent

results were obtained i.e., beta band showed its increase in mental arithmetic task but not in resting task. In Table 4

and Table 5 EEG Band Ratios performances in L/R and R/L ratio of Fp1, Fp2, F3, F4 electrodes for resting task and

mental arithmetic task. In both tables increase in beta and theta, decrease in alpha bands shows the effect of

cognitive load on the subjects while performing the mental arithmetic task, where in resting task it doesn’t show that

effect. Finally, we conclude that EEG Band Ratios in L/R, R/L and (L-R)/(L+R) ratios are one form of techniques

for assessing the cognitive load.

6. Conclusion

Our framework differs from other approaches described in the literature by being the first to use this type of

implementation to handle EEG signals. This paper analyzes the cognitive load from different EEG band ratios using

“bior4.4” decomposition method, 6-pair of electrodes from 6 brain lobe areas and L/R, R/L and (L-R)/(L+R) ratios

of pre-frontal and frontal electrode analysis with EEG band ratios are the first time discussed in this paper.

Classification and EEE band ratios used to assess the variations in lobes (i.e., pre-frontal, frontal, parietal, central,

temporal and occipital) and variations in band frequencies which shows the effect of load on the subjects
performance in both task (i.e., resting task and mental arithmetic task). By applying 4 features (band power, energy,

mean and relative energy) to the classifier, in individual brain lobe areas it is observed that frontal and parietal brain

lobe areas have achieved high accuracy in classifying, whereas central brain lobe areas have achieved low accuracy

in classifying. Pair-wise combinations are implemented to observe the changes in classification results. We observe

that the 3-pair combination has shown a better performance, while 5-pair combination showed less performance. In

[43] EEG band ratios are analyzed for frontal lobe area by considering average band power as feature, subjects

performance is evaluated for pre and post meditation and observed that effect of meditation on cognitive load. Better

improvement observed from pre to post meditation. In our study, EEG band ratios were analyzed by considering the

left and right hemisphere of prefrontal and frontal electrodes with L/R, R/L and (L-R)/(L+R) ratios for assessing the

cognitive load on the brain. Increase in beta observed to be directly proportional to increase in difficulty of the task

and increase in cognitive load. From EEG band ratios we predicted that, prefrontal lobe area in R/L ratio has shown

increase in cognitive load when compared with other ratios, as L/R ratio has also shown the increment but not that

much as R/L. Whereas the ratio (L-R)/(L+R) has not shown that much improvement because of taking the difference

of left and right hemisphere in numerator and summing in denominator. In all L/R, R/L and (L-R)/(L+R) ratios,

increase in beta and decrease in alpha which indicates the cognitive load on the subjects when performing mental

arithmetic task.

ACKNOWLEDGEMENT

This research work was supported by Ministry of Human Resource Development, Govt. of India under Quality

Improvement Program scheme implemented by AICTE. We would like to thanks to Taras Shevchenko National

University of Kyiv, Ukraine (Educational and Scientific Centre “Institute of Biology and Medicine” and Faculty of

psychology) for providing dataset.

Conflict of Interest: The authors declare no conflicts of interest.

REFERENCES
1. Kralikova, I., Babusiak, B., & Smondrk, M. (2022). EEG-Based Person Identification during escalating
cognitive load. Sensors, 22(19), 7154.
2. Atkinson, R. and Shiffrin, R., 1968. Human Memory: A Proposed System and its Control
Processes. Psychology of Learning and Motivation, (Vol. 2, pp. 89-195). Academic Press.
3. Freeman, W. J. (2002). Making sense of brain waves: the most baffling frontier in
neuroscience. Biocomputing, 1-23.
4. Cabañero, L., Hervás, R., González, I., Fontecha, J., Mondéjar, T., & Bravo, J. (2019). Analysis of
cognitive load using EEG when interacting with mobile devices. Multidisciplinary Digital Publishing
Institute Proceedings, 31(1), 70.
5. Sweller, J. (1994). Cognitive load theory, learning difficulty, and instructional design. Learning and
instruction, 4(4), 295-312.
6. Mayer, R., & Mayer, R. E. (Eds.). (2005). The Cambridge handbook of multimedia learning. Cambridge
university press.
7. Clark, R. C., & Mayer, R. E. (2016). E-learning and the science of instruction: Proven guidelines for
consumers and designers of multimedia learning. john Wiley & sons.
8. Klimesch, W. (1999). EEG alpha and theta oscillations reflect cognitive and memory performance: a
review and analysis. Brain research reviews, 29(2-3), 169-195.
9. Anderson, E. W., Potter, K. C., Matzen, L. E., Shepherd, J. F., Preston, G. A., & Silva, C. T. (2011, June).
A user study of visualization effectiveness using EEG and cognitive load. In Computer graphics
forum (Vol. 30, No. 3, pp. 791-800). Oxford, UK: Blackwell Publishing Ltd.
10. Paas, F., Renkl, A., & Sweller, J. (2003). Cognitive load theory and instructional design: Recent
developments. Educational psychologist, 38(1), 1-4.
11. Chalmers, P. A. (2003). The role of cognitive theory in human–computer interface. Computers in human
behavior, 19(5), 593-607.
12. Hart, S. G., & Staveland, L. E. (1988). Development of NASA-TLX (Task Load Index): Results of
empirical and theoretical research. In Advances in psychology (Vol. 52, pp. 139-183). North-Holland.
13. Fink, A., Grabner, R. H., Neuper, C., & Neubauer, A. C. (2005). EEG alpha band dissociation with
increasing task demands. Cognitive brain research, 24(2), 252-259.

14. Pribram K. H. (2013). Brain and perception: Holonomy and structure in figural processing. Psychology
Press.
15. Huang, R. S., Jung, T. P., & Makeig, S. (2009). Tonic changes in EEG power spectra during simulated
driving. In Foundations of Augmented Cognition. Neuroergonomics and Operational Neuroscience: 5th
International Conference, FAC 2009 Held as Part of HCI International 2009 San Diego, CA, USA, July
19-24, 2009 Proceedings 5 (pp. 394-403). Springer Berlin Heidelberg.
16. Antonenko, P., Paas, F., Grabner, R., & Van Gog, T. (2010). Using electroencephalography to measure
cognitive load. Educational psychology review, 22, 425-438.
17. Gevins, A., Smith, M. E., Leong, H., McEvoy, L., Whitfield, S., Du, R., & Rush, G. (1998). Monitoring
working memory load during computer-based tasks with EEG pattern recognition methods. Human
factors, 40(1), 79-91.
18. Delorme, A., & Makeig, S. (2004). EEGLAB: an open source toolbox for analysis of single-trial EEG
dynamics including independent component analysis. Journal of neuroscience methods, 134(1), 9-21.
19. Brouwer, A. M., Hogervorst, M. A., Van Erp, J. B., Heffelaar, T., Zimmerman, P. H., & Oostenveld, R.
(2012). Estimating workload using EEG spectral power and ERPs in the n-back task. Journal of neural
engineering, 9(4), 045008.
20. Cajochen, C., Brunner, D. P., Krauchi, K., Graw, P., & Wirz-Justice, A. (1995). Power density in
theta/alpha frequencies of the waking EEG progressively increases during sustained
wakefulness. Sleep, 18(10), 890-894.
21. Borghini, G., Astolfi, L., Vecchiato, G., Mattia, D., & Babiloni, F. (2014). Measuring neurophysiological
signals in aircraft pilots and car drivers for the assessment of mental workload, fatigue and
drowsiness. Neuroscience & Biobehavioral Reviews, 44, 58-75.
22. Eoh, H. J., Chung, M. K., & Kim, S. H. (2005). Electroencephalographic study of drowsiness in simulated
driving with sleep deprivation. International Journal of Industrial Ergonomics, 35(4), 307-320.
23. Jap, B. T., Lal, S., Fischer, P., & Bekiaris, E. (2009). Using EEG spectral components to assess algorithms
for detecting fatigue. Expert Systems with Applications, 36(2), 2352-2359.
24. da Silveira, T. L., Kozakevicius, A. J., & Rodrigues, C. R. (2016). Automated drowsiness detection through
wavelet packet analysis of a single EEG channel. Expert Systems with Applications, 55, 559-565.
25. Knyazev, G. G., Slobodskoj-Plusnin, J. Y., Bocharov, A. V., & Pylkova, L. V. (2011). The default mode
network and EEG alpha oscillations: an independent component analysis. Brain research, 1402, 67-79.
26. Onton, J., Delorme, A., & Makeig, S. (2005). Frontal midline EEG dynamics during working
memory. Neuroimage, 27(2), 341-356.
27. De Smedt, B., Janssen, R., Bouwens, K., Verschaffel, L., Boets, B., & Ghesquière, P. (2009). Working
memory and individual differences in mathematics achievement: A longitudinal study from first grade to
second grade. Journal of experimental child psychology, 103(2), 186-201.
28. Doppelmayr, M., Klimesch, W., Sauseng, P., Hödlmoser, K., Stadler, W., & Hanslmayr, S. (2005).
Intelligence related differences in EEG-bandpower. Neuroscience Letters, 381(3), 309-313.
29. Zyma, I., Tukaev, S., Seleznov, I., Kiyono, K., Popov, A., Chernykh, M., & Shpenkov, O. (2019).
Electroencephalograms during mental arithmetic task performance. Data, 4(1), 14.

30. Bassam N. A., Ramachandran V., & Parameswaran S. E. (2021). Wavelet Theory and Application in
Communication and Signal Processing. In (Ed.), Wavelet Theory. IntechOpen.
31. Al-Fahoum, A. S., & Al-Fraihat, A. A. (2014). Methods of EEG signal features extraction using linear
analysis in frequency and time-frequency domains. International Scholarly Research Notices, 2014.
32. Sifuzzaman, M., Islam, M. R., & Ali, M. Z. (2009). Application of wavelet transform and its advantages
compared to Fourier transform.
33. Ezra, Y. B., Lembrikov, B. I., Schwartz, M., & Zarkovsky, S. (2018). Applications of wavelet transforms to
the analysis of superoscillations. Wavelet Theory and Its Applications, 195.
34. Sonal, D. K. (2007). A study of various image compression techniques. COIT, RIMT-IET. Hisar, 8, 97-102.
35. Singh, P., Singh, P., & Sharma, R. K. (2011). JPEG image compression based on biorthogonal, coiflets and
daubechies wavelet families. International Journal of Computer Applications, 13(1), 1-7.
36. Salem, M. A., Ghamry, N., & Meffert, B. (2009). Daubechies versus biorthogonal wavelets for moving
object detection in traffic monitoring systems. Humboldt-Universität zu Berlin, Mathematisch-
Naturwissenschaftliche Fakultät II, Institut für Informatik.
37. Sridhar, S., Kumar, P. R., & Ramanaiah, K. V. (2014). Wavelet transform techniques for image
compression-an evaluation. International journal of image, graphics and signal processing, 6(2), 54.
38. Prasad, P. M. K., Prasad, D. Y. V., & Rao, G. S. (2016). Performance analysis of orthogonal and
biorthogonal wavelets for edge detection of X-ray images. Procedia Computer Science, 87, 116-121.
39. Mallat, S. (1999). A wavelet tour of signal processing. Elsevier.
40. Gupta, S. S., Taori, T. J., Ladekar, M. Y., Manthalkar, R. R., Gajre, S. S., & Joshi, Y. V. (2021).
Classification of cross task cognitive workload using deep recurrent network with modelling of temporal
dynamics. Biomedical Signal Processing and Control, 70, 103070.
41. Nagendra, H., Kumar, V., & Mukherjee, S. (2015). Cognitive behavior evaluation based on physiological
parameters among young healthy subjects with yoga as intervention. Computational and mathematical
methods in medicine, 2015.
42. Travis, F., & Shear, J. (2010). Focused attention, open monitoring and automatic self-transcending:
categories to organize meditations from Vedic, Buddhist and Chinese traditions. Consciousness and
cognition, 19(4), 1110-1118.
43. Jadhav, N., Manthalkar, R., & Joshi, Y. (2017, June). Assessing effect of meditation on cognitive workload
using EEG signals. In Second International Workshop on Pattern Recognition (Vol. 10443, pp. 269-273).
SPIE.

You might also like