Brain computer interface for measuring the impact of yoga on concentration levels in engineering students

Abstract

Background:

Students have to manage the strain of rising education level and their future career, accompanying the hormonal changes during their pubescence. This creates a great impact on their education as well as personal life. In this paper, an analysis has been made to study the impact of yoga on engineering students. To understand the impact. Brain-Computer Interface (BCI) approaches have been utilized. An EEG based BCI is used which will give a direct view of whats going on in the students’ brains.

Methodology:

In this work, an experiment has been performed on engineering students and their brain activity is recorded before and after practicing yoga. In the experimental procedure, EEG signals are acquired from 8 electrodes which are associated with the cognitive and memory-related tasks of the brain. During each trial, participants solve the set of mathematical questionnaire. EEG signals are acquired during test trials before and after the yoga session. A bandpass filter is applied to preprocess the EEG signals. A discrete wavelet transform is implemented for feature extraction of the preprocessed signals.

Results:

Different classification algorithms are applied to classify the EEG signals before and after the yoga session. To measure the classification performance, measures such as accuracy, sensitivity, and specificity are presented in the paper. The highest accuracy of 95 % is achieved with Probabilistic Neural Network. Classification concluded the variations in signals before and after yoga. Further, in this work analysis of frequency bands, accuracy and score of the subjects before and after the yoga session are also done.

Keywords

Brain Computer Interface EEG signals yoga wavelets Probabilistic Neural Network

1 Introduction

Brain-Computer Interface (BCI) is a communication technology that enables communication between the human brain and different devices without physical contact through the generation of brain signals. BCI setup consists of an amplifier, sensing device and a monitoring system. In a BCI application, there are three stages: signal acquisition (capturing the brain signals), pre-processing (noise removal) the signals and classification. Since the invention of Electroencephalography (EEG), researchers are trying to get meaningful information from brain signals. There are also different methods for signal acquisition which includes fMRI (Functional Magnetic Resonance Imaging) [29], NIRS (Near-Infrared Spectroscopy) [30], Magnetoencephalography (MEG) [31], optical coherence tomography (OCT) [32 –35] etc. During the early time, BCI is mostly used for the provision of helpful technologies to the disabled person. BCI has a wide number of applications and has proven to be most beneficial, which includes applications from controlling the cursor of a computer through commands of mind to recognizing the emotions of a person. One of the vital applications of BCI is recognizing the stress levels in an individual by analyzing the brain waves [1]. This has proved beneficial, as understanding the stress level at an early stage with good accuracy can help the individual to implement proper stress management techniques. Another approach is detecting the mental concentration of the player during trianing [2]. The trainer can analyze the external factors that affect the concentration of the player so that he can provide the training to the player more specifically. In this, brain waves are considered for analyzing the mental condition. Different types of waves are generated by brain, which are delta(δ), theta(θ), alpha(α), beta(β), and gamma(γ). Each wave has a different frequency and the presence of each frequency depends on the activities performed by an individual.

2 Related work

In today’s era, students have to manage the burden of growing education levels and increasing competition. Including this, the entertainment world and social media have attained their height and it is also proving to be one of the reasons for distraction. When students are bewildered and stressed out, they breathe in an uncomfortable way. Their breathing is protracted for a long time than the normal inspiration, thus their body gets accumulated with the toxins, which disturbs their mental state [6]. This unmanaged stress results in the development of mental and behavioral problems in students, which leads to anxiety, depression and lack of focus [7]. Certain researches have revealed that 13.1% of students suffer from severe, 37.7% of students suffer from modest and of students suffer from 2.4% extremely severe depression [5]. These factors affect the concentration of power students as well as their psychic stability. Lack of concentration results in the breakdown of students, achieving a high level of concentration is the key to success.: Yoga,the most ancient Indian exercising technique, that consists of different poses (asana), meditation (dhyana), breathing (pranayama), relaxation of the senses, concentration, attaining a higher level of consciousness is believed to bring stability and health to the mind and body [4]. This paper includes an experiment that is based on how yoga increases the concentration of the students and this analysis is done using BCI.

3 Methodology

The architecture of the experiment is stated in Fig. 1.

Fig. 1

Architecture of Experiment.

3.1 Data collection

EEG signals are acquired from 10 subjects. Brain Vision Recorder is utilized for signal acquisition. The subjects had a normal vision and they did not have any neurological disorder so that it should not deviate the results of this experiment. The subjects were associated with the age group of 12-22 years. The subjects were preferred of this age group because they covered school children, teenagers as well as adults. School children were chosen because they are at the beginning of puberty. Teenagers were chosen because they undergo a lot of hormonal changes and adults were chosen because they suffer from stress-related to their career. Fp1, Fp2, F3, F4, T7, T8, C3, C4 electrodes were selected for data acquisition. Fp1 and Fp2 electrodes relate to the decision making part of the brain. F3, F4, C3, C4 electrodes relate to concentration and memory part of the brain. T7 and T8 belong to the processing semantics. Table 1 explains the functions of the brain in various regions with respect to the electrodes. The experimental setup is illustrated in Fig. 2.

Table 1
Functioning of the brain in various regions with respect to the electrodes

Electrodes Brain region Function

Fp1,Fp2 Prefrontal part of frontal lobe Decision making, Planning.

F3,F4 Frontal lobe Attention, Memory, Thinking.

T7,T8 Temporal lobe Processing semantics in both vision and speech.

C3,C4 Frontal lobe(Primary motor cortex and near central sulcus) Memory, Thinking.

Electrodes	Brain region	Function
Fp1,Fp2	Prefrontal part of frontal lobe	Decision making, Planning.
F3,F4	Frontal lobe	Attention, Memory, Thinking.
T7,T8	Temporal lobe	Processing semantics in both vision and speech.
C3,C4	Frontal lobe(Primary motor cortex and near central sulcus)	Memory, Thinking.

Fig. 2

Experimental Setup.

Before proceeding to the experiment, subjects are delivered with an elaborative introduction and explanation about the experiment. Experiment procedure has been divided into three phases: Problem-solving before yoga, Yoga, Problem-solving after yoga. In the first phase, subjects were asked to solve a set of mathematical questions. Questions are set according to the education level and age of the subject. Time for solving questions is 40 seconds. EEG signals are recorded during that period with the sampling rate of 250 Hz. Five trials are conducted for each subject for 5 consecutive days, as the mental state of the subject is different every day. In the second phase, subjects are instructed to practice yoga for two weeks for 20 minutes per day. Yoga practice for 20 days included physical poses as well as breathing exercises. In the third phase, after two weeks, before conducting each test, subjects are asked to perform yoga for 5 minutes and then they are asked to solve the set of mathematical questions for 40 seconds.

3.2 Data preprocessing

Although, EEG device helps to record the signals of the brain activity, it also records the signals from the sources other than the brain. These extra signals recorded, are known as the artifacts. EEG signals are accordingly preprocessed to remove these artifacts and to acquire precise signals. In this experiment, a bandpass filter is applied to the data. Bandpass filter permits the signals at a certain range of frequency which form a band of frequency and rejects other frequencies. Bandpass filter used in this experiment filtered the frequency in the range of 0.3-30 Hz, which eliminates the high-frequency bands from the data.

3.3 Feature extraction

Preprocessed signals were used for feature extraction. Analog EEG signals of all subjects are converted to numerical data i.e into (.)mat file using the Brain Vision Analyzer. The matrix of data for each subject was obtained for a single trial. Now, Discrete Wavelet transform is applied to the data of each subject at 4 levels, using ‘Symlet 4’ as a mother wavelet. Numerical value 4 in Symlet 4 indicates four vanishing moments. Symlet is the modification of Daubechies Wavelets. The scaling function of Symlet wavelet integrates to unity. EEG signals are non-stationary waves in which the frequency shifts at different times. Therefore it is very important to have time-frequency representation of the EEG. signals. Following techniques are used for the feature extraction of the EEG signals: Hilbert-Huang Transform (HHT), Principal Component Analysis (PCA), Local Discriminant Basis, Wavelet transform, Independent Component Analysis (ICA) [4]. A wavelet transform provides time (t) and frequency (f) relation simultaneously, thus it gives the time-frequency representation of the signals. Wavelet transform overcomes the resolution related problem of STFT. So Wavelet transform was used in this study. Discrete wavelet transform (DWT) decomposes the signals into orthogonal sets of wavelets. EEG signals are decomposed simultaneously using the low pass filter and high pass filter. The signals are decomposed up to the required level. This process is called Downsampling of the signals. The decomposed wavelet coefficients comprise the information of original signal [26]. Now, half of the data can be eliminated using the Nyquist’s rule. Haar wavelet was the first DWT wavelet discovered by Hungarian mathematician Alfred Haar [27]. Other widely used DWT wavelets include Daubechies wavelets, Coiflet Wavelets, Symlet Wavelets, etc. Daubechies is a wavelet of order 4, Coiflet is a wavelet of order 3, Symlet is a wavelet of order 8. The order defines the smoothness of the wavelet. Therefore Symlet Discrete wavelet transform was used for feature Extraction.

In the first level of signal decomposition, the signal was decomposed to high-frequency components and the low-frequency components as illustrated in Fig. 3. Here, high-frequency components comprise of approximate coefficients and low-frequency components comprises of detail coefficients. In the second level, the detail coefficients are further decomposed to approximate and detail coefficients. This process is repeated until level four. It can be observed in Fig. 3, D2, D3, D4 are the detail coefficients at levels two, three and four respectively. In this study, D2, D3, D4 were reconstructed again to obtain the extracted features of data. Figure 4 illustrates the four decomposed levels of feature extraction of this experiment.

Fig. 3

Process of feature extraction using Discrete Wavelet transform.

Fig. 4

EEG signals decomposition using DWT upto 4 levels.

3.4 Classification

The extracted features from wavelets are used for classification. The data is classified into two classes: concentration before yoga and concentration after yoga. Classification algorithms work uniquely on different datasets i.e A classifier can work productively for one dataset but it can work unproductive for another dataset. Different classifiers are implemented on the dataset and an optimum model is adopted. The implemented classifiers includes Support Vector Machine (SVM), K-Nearest Neighbor (k-NN) and Probabilistic Neural Networks (PNN).

1) SVM: SVM is considered to be a superior signal classifier based on pattern recognition [13]. The study specified by Mervyn V. M. Yeo [13] concluded that detection of EEG changes is possible by SVM. SVM is useful for classification of data into two classes and it is a supervised learning algorithm. Multiclass SVM is used for classification of data into more than two classes [12]. In SVM, data is plotted in N-dimensional space (N=no. of features). It determines hyperplane which maximizes the margin. It has kernels, it transforms a non-separable problem into a separable problem. It also has other tuning parameters named gamma, regularization and margin parameters. In this study, rbf kernel was used to achieve high accuracy and gamma and regularization parameters were tuned. 2) k-NN: k-NN can be used for classification as well as regression. In this study, k-NN is used for classification. It is a non-linear classifier. It assigns a given input sample to class, based on majority class from k-nearest training samples. The class assignment in k-NN has two approaches, that is, similarity voting and majority voting [14]. In this experiment, majority voting is used as class assignment approach. This approach is effective in this experiment, because it is less susceptible to outliers. Optimum value of k was achieved after using pipeline. Accuracy of this algorithm on data was 70

3) Probabilistic Neural Networks: Probabilistic Neural Networks [36] is a variant of artificial neural network, where probability density function is calculated for each class. Using probability density function of each class, the class probability of a new input data is estimated and Bayes rule is employed to allocate the class with highest posterior probability to new input data. This classification approach minimizes the mis-classification rate. In most of the Neural Networks, the activation function used is a sigmoid function. In PNN sigmoid function is substituted with exponential function. PNN has four layers namely input layer (no. of neurons= no. of features), the pattern layer (Parzen Window density estimate), the summation layer (no. of neurons=no. of classes) and the output layer. Input layer transfers the input data to each neuron of the pattern layer. In Fig. 1, input vector x is passed to the pattern layer p and further to the summation layer c and finally, in the output layer, the class of the input vector is calculated. The probability distribution function (PDF) of the test input vector is evaluated by the pattern layer for each class by applying the Parzen Window Density Estimate method. Parzen Window does not include parameters and PDF of a given input is estimated by the superposition of the number of functions (particularly Gaussian function).

In this experiment input to the PNN are 8 electrode’s data, that is the input layer has 8 neurons. Then the data of 8 electrodes is transferred to the pattern layer. The computation of the pattern layer is performed with the Parzen Window Density estimate. Then the output of the pattern layer is further passed to the summation layer. Summation layer pass its output as input to the ouput layer, which apply Bayes rule to find class of input vector. The classification was binary that is whether that point belongs to before yoga or after yoga class. Among PNN, SVM, k-NN, highest accuracy is observed in PNN, which is referred in Table 2.

Table 2
Simulations of classification algorithms of EEG data

Algorithm Accuracy Sensitivity Specificity

PNN 95% 96% 94%

KNN 70% 80% 82%

SVM 90% 85% 88%

Algorithm	Accuracy	Sensitivity	Specificity
PNN	95%	96%	94%
KNN	70%	80%	82%
SVM	90%	85%	88%

The extracted features were used for classification. The data was classified into two classes: concentration before yoga and concentration after yoga. The data was divided into two groups. First was before yoga which was labeled as A and second one was after yoga which was labeled as B. We have used these two classes for classification.Data was randomly shuffled and divided 80:20 portion for training testing. Classification was done to show that data was distinguishable before and after yoga. It indicated that there was change in brain signals after performing yoga. We used classification algorithms to predict whether particular data point to belong class A or class B. Classification algorithms work uniquely on different datasets i.e A classifier can work productively for one dataset but it can work unproductive for another dataset. Different classifiers were implemented on the dataset and an optimum model was adopted. The implemented classifiers includes Support Vector Machine(SVM), K-Nearest Neighbour (k-NN) and Probabilistic Neural Networks (PNN).

4 Experimental analysis

In classification, it is observed that there is a change in EEG data before and after yoga. Therefore, analysis of the signals is performed to examine impact of yoga on EEG signals. This section is broadly divided into two parts. The first part is based on an analysis of the score and accuracy of the conducted test. The second part is based on the analysis of EEG signals concerning alpha and beta waves. After the experiment, a survey was conducted, to check the consistency of the Yoga performance of the students. The response of students about their experience before and after yoga was noted at the end of the experiment. From responses, it was observed that subject S5 was irregular and other subjects were regular. The irregularity of the subject S5 has also affected his score as well as alpha and beta band power. The regularity of the subjects is reflected in their score as well as alpha and beta band power.

4.1 Analysis of score and accuracy

For experiment and analysis, questions are set according to the education level and age of the subject. For every correct answer, 250 points are rewarded while for the incorrect answer 125 points are subtracted from the total score. The accuracy obtained in table 3 is not exclusively reliant on the score. It is considered as a measure to analyze the impact of yoga because every subject has different excellency in mathematics. Accuracy enables the analysis of subjects at the same level irrespective of their age and education level. It is computed by the following formula: $Accuracy = \frac{C_{1} + C_{2} + C_{3} + C_{4} + C_{5}}{T_{A}}$ (1) In equation 1, C is the number of correct attempts of the subject in each trial and T _A is the total number of questions attempted by the subject in all 5 trials.

Accuracy is a measure of concentration as well as memory tasks in this test as the subjects are asked to solve the questions without using pen and paper. It has been observed that in case of every subject there is a significant change in accuracy besides the subject S5. After feedback from the subject S5, it is noted that the S5 is irregular in performing yoga. Subjects S1 and S4 have high improvement in accuracy and average marks among all subjects. During feedback from those subjects, they revealed that yoga has helped them to manage their anxiety as well as distractions. They also felt an improvement in their concentration and alertness while performing different tasks. Subjects S2, S3, S6, S7, S8, S9, and S10 also exhibited better improvement in accuracy and score. Yoga has marked propitious in the performance of response time with respect to cognitive process [17]. It is observed in Table 3 that average score of every subject after yoga is improved. The time for solving mathematical question before and after remained the same but it is discovered that score is increased, eventually, the number of questions endeavored by students are increased after yoga. It indicates that their response time to solve the questions is reduced.

Table 3

Subject wise variation in accuracy and score

Subject	Before		After
	Avg score	Avg accuracy	Avg score	Avg accuracy
S1	1460	79 %	2275	92 %
S2	1650	84 %	1825	88 %
S3	1812	81 %	1975	84 %
S4	1685	83 %	2565	93 %
S5	1245	71 %	1300	73 %
S6	1725	81 %	1855	85 %
S7	1500	85 %	1650	89 %
S8	1645	82 %	1900	87 %
S9	1200	70 %	1475	75 %
S10	1250	72 %	1400	76 %

4.2 EEG analysis

The brain waves acquired from the subjects are analyzed. The principal focus in this work is on the alpha band and the beta band. Alpha and beta wave lie in range of 8-13Hz and 13 to 30 Hz respectively. The alpha activity is responsible for attention and memory-related task whereas beta activity is responsible for cognitive processes [16]. This section is subdivided into three sections. The first section deals with an analysis of the alpha band. The second section deals with the beta band and third one with the intrabeta band activity. This analysis is based on the power of the frequency bands with respect to Power Spectral Density (PSD). Power Spectral Density is obtained with the help of pwelch() function in MATLAB, using the EEGlab tool for the data of each subject.

4.2.1 Analysis of alpha band power

The alpha band extends in the frequency range of 8-13 Hz. The alpha band is associated with memory, relax and calm state and consciousness of the person. Fast and accurate recalling from memory has a direct correlation with the alpha frequency band [17]. It is also noted that the alpha band is associated with two primary functions of attention: selection and suppression [18]. It enables semantic orientation i.e capability to be consciously oriented in space and time. In the study by Puma, it was observed that alpha band power is used for assessing the cognitive workload in a different multitasking environment [19].

For analyzing the alpha band power, it is necessary to select the electrodes which have high alpha power, therefore channel spectra and maps were plotted using spectopo() function. Function spectopo() plots the power spectral density (PSD) in a given frequency, it also incorporates the respective electrodes which have the presence of high power as well as low power. The red color on the map denotes the highest power whereas the blue color denotes lowest power. After plotting the channel spectra and maps for the alpha frequency, it was observed that the PSD of 60 % of the subjects are in the frontal and prefrontal region and 40 % is observed to be in the temporal region. Figure 5 represents the above explanation of power distribution. As the power distribution included every electrode, all the electrodes are selected for the analysis of alpha-band power. For the analysis of the alpha band power, before and after yoga, the power of each electrode has been calculated for each trial of the subject using pwelch() function. Then the power of all electrodes has been summed and subsequently, the average of each trial is calculated. Further, the result of every trial is summed and subsequently average of before and after trials is calculated. Table 4 represents the alpha power before and after yoga, in terms of decibel (dB).

Fig. 5

Power Spectral Density distribution of alpha band.

From Table 4, it is observed that all subjects have good improvement in their alpha power except subject S5 with an improvement of only 1%, the same result was obtained in Table 3. Even though subject S8 has an accuracy improvement of 5%, the alpha power improvement is very high corresponding to other subjects, which defines that improvement in the alertness of subject S8 is high. Subject S4 has an improvement in accuracy by 10%, which is precisely correlated with the high improvement of alpha-band power. The other subjects S1, S2, S3, S6, S7, S9, S10 had also recorded with a significant change in alpha band power, which reflects that yoga has availed them in the improvement of consciousness and memory-related tasks.

Table 4

Discrepancy in Alpha band power

Subject	Alpha band power		Change
	Before	After
S1	1.7541	1.9113	8% ↑
S2	2.0774	2.2302	7% ↑
S3	1.7445	1.9336	11% ↑
S4	0.9544	1.1385	19% ↑
S5	0.6810	0.6890	1% ↑
S6	1.3278	1.5160	14% ↑
S7	1.6543	1.7246	4% ↑
S8	0.9257	1.1625	25% ↑
S9	2.0587	2.3624	14% ↑
S10	1.0258	1.1325	10% ↑

4.2.2 Analysis of beta band power

The beta band extends in the frequency range of 13-30 Hz. Beta waves define the characteristics of the engaged mind [20]. Beta waves are also effective for active response [21]. A study defines that an increase of existing beta wave activity can lead to a higher GPA among students [15]. It has been also observed that there is an improvement in the beta band at the prefrontal region during the problem-solving task as compared to the resting state [22]. According to researchers, theta wave decreases and beta wave increases during concentration and the power ratio of beta and theta bands were used for determining the concentrated state of a person [23].

4.2.3 Analysis of intrabeta band

Analysis of complete Beta band power

Channel spectra and maps are plotted for the beta band power distribution using spectopo() function, to obtain the electrodes with high beta power. In overall observation, the power distribution of beta band was 50% in the right hemisphere and 50 % in the left hemisphere of the subjects, which comprise all 8 electrodes. Refer Fig. 6 for the power distribution of the beta band.

Fig. 6

Power Spectral Density of beta band.

The process of power calculation for the beta band in terms of PSD was repeated similar to the process of computing the power of alpha using pwelch() function. The power of the beta band was estimated for each subject, before and after yoga. Table 6 outlines the beta power before and after yoga, in terms of decibel (dB).

Stipacek in his paper noted that, as memory load raises, accuracy decreases and response time increases [24]. But in this experiment, it was noticed that even though memory load is increased accuracy increases and response time decreases after going through the yoga session. It indicates that even though memory load is increased the cognitive power of the human mind gets benefitted from yoga. It is observed from Table 3 that after the yoga session, students were capable of handling higher memory load and they were also able to accurately calculate mathematical questions as compared to test before the yoga session. That is their memory load was increased but the level of the cognitive task carried out by them was improved. The same thing is observed in Table 6, there is an improvement in the beta power of every subject which is associated with the cognitive task of the human.

From Table 6, it is observed that subject S4 has drastic growth in beta power, which is also reflected in Table 3. Even though subject S5 is irregular, there is an improvement in its beta band power, which indicate that subject S5 had an enhancement in concentration. Subjects other than S4 and S5 has also shown improvement in the beta power. During the feedback from all the subjects, it is discovered that the subjects are able to solve the given set of questionnaire more quickly and with good concentration after yoga.

Analysis of subparts of beta band. The increase in beta band power indicates that there is an improvement in the cognitive activity of human. But it does not indicate explicitly, which cognitive activity of the human is increased. To deal with this issue, the beta band is further divided into three sub-parts as given in Table 7.

Table 7 shows that improvement in low and mid beta band is beneficial, whereas improvement in high beta band is disadvantageous. Therefore, it is necessary that when beta band power increases, the power of high beta band should not increase. Table 5 precisely indicates that there is a positive impact on the power of low beta and mid beta, whereas in case of high beta it has a negative relationship. Even though subject S5 was irregular it shows the same trend.

Table 5

Discrepancy in Beta band power

Subject	Beta band power		Change
	Before	After
S1	0.5831	0.6233	7% ↑
S2	0.7452	0.8706	17% ↑
S3	0.3897	0.4097	5% ↑
S4	0.3333	0.5010	50% ↑
S5	0.1877	0.2080	11% ↑
S6	0.6528	0.7922	21% ↑
S7	0.7992	0.9164	15% ↑
S8	0.5711	0.6670	17% ↑
S9	0.3945	0.4472	13% ↑
S10	0.6978	0.8123	16% ↑

Table 6

Intrabeta waves and their properties

Wave	Frequency(Hz)	Activities
Low beta wave	13-15	Focus, Engaged activity
Mid beta wave	16-20	Thinking, Consciousness
High beta wave	21-30	Tension, Stress, Excitement

Table 7

Discrepancy in Intrabeta band power

Subject	Low Beta band			Medium Beta band			High Beta band
	Before	After	Change	Before	After	Change	Before	Before	Change
S1	0.8746	0.9279	6% ↑	0.6053	0.7271	20% ↑	0.5113	0.4837	5% ↓
S2	1.3087	1.3441	3% ↑	0.9390	1.0806	15% ↑	0.5941	0.4173	30% ↓
S3	0.7641	0.8217	8% ↑	0.4695	0.5165	10% ↑	0.2930	0.2671	9% ↓
S4	0.3584	0.5259	35% ↑	0.4130	0.5184	26% ↑	0.4895	0.3164	35% ↓
S5	0.2546	0.2960	16% ↑	0.2371	0.2447	3% ↑	0.1996	0.1274	10% ↓
S6	0.7356	0.8165	11% ↑	0.6012	0.6720	12% ↑	0.2720	0.2516	8% ↓
S7	0.5378	0.6275	17% ↑	0.4223	0.5122	21% ↑	0.4769	0.3990	16% ↓
S8	1.1752	1.2700	8% ↑	1.0238	1.1205	9% ↑	0.4892	0.3842	21% ↓
S9	0.4121	0.4927	20% ↑	0.3578	0.4432	23% ↑	0.3556	0.2960	17% ↓
S10	0.6358	0.7214	13% ↑	0.6819	0.7720	13% ↑	0.3516	0.3125	11% ↓

Figure 7 distinctly shows this fact for single trial. Power of low and mid beta is high in after yoga, whereas low in before yoga. In case of high beta, it has been observed that overall power is depleted after yoga.

Fig. 7

Variation in Activity Power Spectrum beta band.

5 Conclusion

This work is focused on the study of the impact and effectiveness of yoga on engineering students. The experiment is conducted in three phases: test trials before the yoga session, 20 days of yoga practice, test trials after the yoga session. Further, EEG signals are classified and analyzed to determine its positive impact on cognitive and memory-related task. Throughout data acquisition, subjects solved the set of mathematical questionnaire. The signals are recorded during the tests. For classifying these EEG signals, three different classifiers are applied. The highest accuracy of 95 % is achieved with the Probabilistic Neural Network. Further, to distinguish the positive or negative impact of yoga on the students, analysis of the accuracy and score as well as frequency bands is performed. Accuracy and score analysis concluded that the accuracy of the subjects in solving mathematics is increased after yoga. Alpha and beta band power analysis with respect to Power Spectral Density is performed. Alpha band power analysis concluded that the power of the alpha band increases after yoga which results in the improvement of handling memory-related tasks by the subject. Beta band power analysis concluded that the power of the beta band increases after yoga which results in improvement of the cognitive ability of the students. Also, there is an increment in low and mid beta power which indicates the improvement in concentration and there is a drop in high beta power which indicates the reduction in stress. Further, in course of analysis of feedback from students, it is observed that subjects experienced academics less stressful as compared to before yoga sessions. Along with this, they felt less anxiety during exams and able to cope up with the pressure when they continued Yoga. It indicates that Yoga helps in cognitive activity and memory-related tasks of students and it also helps to reduce the personal and academic stress. Hence, it can be inferred that it is beneficial to include yoga in the curriculum of engineering students.

References

Berger

, Über das Elektrenkephalogramm des Menschen, Archiv fu¨rPsychiatrie undNervenkrankheiten108(3) (1938), 407–431.

Jebelli

, Hwang

and Lee

, EEG-based workers’ stress recognition at construction sites, Automation in Construction93 (2018), 315–324.

Sakai

, Yagi

and Ishii

, EEG analysis of mental concentration in golf putting. In Biomedical Engineering International Conference (BMEiCON), 2012, (pp. 1–5), IEEE.

Azlan

W.A.W.

and Low

Y.F.

, Feature extraction of electroencephalogram (eeg) signal-a review. In Biomedical Engineering and Sciences (IECBES), 2014 IEEE Conference on (pp. 801–806), 2014, December, IEEE.

Ross

, Friedmann

, Bevans

and Thomas

, National survey of yoga practitioners: mental and physical health benefits, Complementary Therapies in Medicine21(4) (2013), 313–323.

Deb

, Banu

P.R.

, Thomas

, Vardhan

R.V.

, Rao

P.T.

and Khawaja

, Depression among Indian university students and its association with perceived university academic environment, living arrangements and personal issues, Asian Journal of Psychiatry23 (2016), 108–117.

Zhaoyuan

, Research into Yoga’s promoting functions for men’s health. In Human Health and Biomedical Engineering (HHBE), 2011 International Conference on (pp. 689–692), 2011, August. IEEE.

Suldo

S.M.

, Shaunessy

and Hardesty

, Relationships among stress, coping, and mental health in high-achieving high school students, Psychology in the Schools45(4) (2008), 273–290.

Mooventhan

and Nivethitha

, Evidence based effects of yoga in neurological disorders, Journal of Clinical Neuroscience43 (2017), 61–67.

10.

Göksel

B.K.

, Coşkun

, Ucler

, Karatas

, Ozge

and Ozkan

, Use of complementary and alternative medicine by a sample of Turkish primary headache patients, Agri26(1) (2014), 1–7.

11.

Adeli

and Panakkat

, A probabilistic neural network for earthquake magnitude prediction, Neural Networks22(7) (2009), 1018–1024.

12.

Specht

D.F.

, Probabilistic neural networks, Neural Networks3(1) (1990), 109–118.

13.

Aravind

K.R.

, Raja

, Mukesh

K.V.

, Aniirudh

, Ashiwin

and Szczepanski

, Disease classification in maize crop using bag of features and multiclass support vector machine, 2018 2nd International Conference on Inventive Systems and Control (ICISC), Coimbatore, India, 2018, pp. 1191–1196.

14.

Yeo

M.V.

, Li

, Shen

and Wilder-Smith

E.P.

, Can SVM be used for automatic EEG detection of drowsiness during car driving?, Safety Science47(1) (2009), 115–124.

15.

Chaovalitwongse

W.A.

, Fan

Y.J.

and Sachdeo

R.C.

, On the time series-nearest neighbor classification of abnormal brain activity, IEEE Transactions on Systems, Man, and Cybernetics-Part A: Systems and Humans37(6) (2007), 1005–1016.

16.

Desai

, Tailor

and Bhatt

, Effects of yoga on brain waves and structural activation: A review, Complementary Therapies in Clinical Practice21(2) (2015), 112–118.

17.

Ray

W.J.

and Cole

H.W.

, EEG alpha activity reflects attentional demands, and beta activity reflects emotional and cognitive processes, Science228(4700) (1985), 750–752.

18.

Sharma

V.K.

, Rajajeyakumar

, Velkumary

, Subramanian

S.K.

, Bhavanani

A.B.

, Madanmohan

A.S.

and Thangavel

, Effect of fast and slow pranayama practice on cognitive functions in healthy volunteers, Journal of Clinical and Diagnostic Research: JCDR8(1) (2014), 10.

19.

Klimesch

, Alpha-band oscillations, attention, and controlled access to stored information, Trends in Cognitive Sciences16(12) (2012), 606–617.

20.

Puma

, Matton

, Paubel

P.V.

, Raufaste

É.

and El-Yagoubi

, Using theta and alpha band power to assess cognitive workload in multitasking environments, International Journal of Psychophysiology123 (2018), 111–120.

21.

Mankar

, Bawane

, Landge

, Dalal

and Gour

, Brain wave sensing. In India Conference (INDICON), 2014 Annual IEEE (pp. 1–5), 2014, December, IEEE.

22.

Dustman

R.E.

, Boswell

R.S.

and Porter

P.B.

, Beta brain waves as an index of alertness, Science137(3529) (1962), 533–534.

23.

Roslan

N.S.

, Amin

H.U.

, Izhar

L.I.

, Saad

M.N.M.

and Sivapalan

, Role of EEG delta and beta oscillations during problem solving tasks. In Intelligent and Advanced Systems (ICIAS), 2016 6th International Conference on (pp. 1–4), 2016, August, IEEE.

24.

Choi

T.J.

, Kim

J.O.

, Jin

S.M.

and Yoon

, Determination of the concentrated state using multiple EEG channels, International Journal of Computer, Electrical, Automation, Control and Information Engineering8(8) (2014), 1373–1376.

25.

Stipacek

, Grabner

R.H.

, Neuper

, Fink

and Neubauer

A.C.

, Sensitivity of human EEG alpha band desynchronization to different working memory components and increasing levels of memory load, Neuroscience Letters353(3) (2003), 193–196.

26.

Bruce

L.M.

, Koger

C.H.

and Li

, Dimensionality reduction of hyperspectral data using discrete wavelet transform feature extraction, IEEE Transactions on Geoscience and Remote Sensing40(10) (2002), 2331–2338.

27.

Haar

, On the theory of orthogonal functional systems, Mathematical Annals69(3) (1910), 331–371.

28.

Bablani

, Edla

D.R.

, Tripathi

and Kuppili

, An efficient Concealed Information Test: EEG feature extraction and ensemble classification for lie identification, Machine Vision and Applications, (2018), 1–20.

29.

Huettel

S.A.

, Allen

W.S.

and Gregory

, Functional magnetic resonance imaging. Vol. 1. Sunderland, MA: Sinauer Associates, 2004.

30.

Villringer

, et al., Near infrared spectroscopy (NIRS): a new tool to study hemodynamic changes during activation of brain function in human adults, Neuroscience Letters154(1-2) (1993), 101–104.

31.

Vrba

and Stephen

E.R.

, Signal processing in magnetoencephalography, Methods25(2) (2001), 249–271.

32.

Men

, et al., Optical coherence tomography for brain imaging and developmental biology, IEEE Journal of Selected Topics in Quantum Electronics22(4) (2015), 120–132.

33.

Ratheesh

K.M.

, Seah

L.K.

and Murukeshan

V.M.

, Spectral phase-based automatic calibration scheme for swept source-based optical coherence tomography systems, Physics in Medicine and Biology61(21) (2016), 7652–7663.

34.

Meleppat

R.K.

, Shearwood