emotionanalysisofcollegestudentsusingafuzzysupport...

Research ArticleEmotion Analysis of College Students Using a Fuzzy SupportVector Machine

Yan Ding 1 Xuemei Chen 2 Shan Zhong 3 and Li Liu 4

1Department of Logistics Support Changshu Institute of Technology Changshu 215500 Jiangsu China2School of Mechanical Engineering Beijing Institute of Technology Beijing 100081 China3School of Computer Science and Engineering Changshu Institute of Technology Changshu 215500 Jiangsu China4Jiangsu Vocational College of Information Technology Wuxi Jiangsu 214153 China

Correspondence should be addressed to Xuemei Chen chenxue781biteducn

Received 26 July 2020 Accepted 24 August 2020 Published 10 September 2020

Guest Editor Yi-Zhang Jiang

Copyright copy 2020 Yan Ding et al is is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

With the rapid development of society the number of college students in our country is on the rise College students are underpressure due to challenges from the society school and family but they cannot find a suitable solution As a result thepsychological problems of college students are diversified and complicated e mental health problem of college students isbecoming more andmore serious which requires urgent attentionis article realizes the monitoring of university mental healthby identifying and analyzing the emotions of college students is article uses EEG to determine the emotional state of collegestudents First feature extraction is performed on different rhythm data of EEG and then a fuzzy support vector machine (FSVM)is used for classification Finally a decision fusion mechanism based on the D-S evidence combination theory is used to fuse theclassification results and output the final emotion recognition results e contribution of this research is mainly in three aspectsOne is the use of multiple features which improves the efficiency of data use the other is the use of a fuzzy support vector machineclassifier with higher noise resistance and the recognition rate of the model is better e third is that the decision fusionmechanism based on the D-S evidence combination theory takes into account the classification results of each feature and theclassification results assist each other and integrate organically e experiment compares emotion recognition based on singlerhythm multirhythm combination and multirhythm fusion e experimental results fully prove that the proposed emotionrecognition method can effectively improve the recognition efficiency It has a good practical value in the emotion recognition ofcollege students

1 Introduction

On contemporary university campuses the number ofcollege students with mental illnesses is increasing day byday Many college students have difficulty adapting tocollege life for a while and a series of mental healthproblems arise which seriously affect their normal studyand life At present the mental health prevention work ofcollege students mainly relies on the inquiry methods ofcounselors and class teachers e problems with thisapproach are as follows (1) e number of teachers is farlower than the number of potential mental health prob-lems In addition the work of university teachers iscomplicated and the workload is huge with limited energy

and time erefore the prevention and treatment ofmental health among college students are often a mereformality (2) At present the main way to carry out mentalhealth prevention and control work is questionnaire sur-vey which is difficult to identify students with real psy-chological problems is has given birth to moreintelligent mental health investigation and preventionmethods e emotional changes of college students canreflect their mental health problems to a certain extent If acollege student is in a sad or neutral state for a long time itindicates that the student has some psychological prob-lems At this time the teacher can start mental healthcounseling in time erefore emotion recognition forcollege students is of great significance

HindawiMathematical Problems in EngineeringVolume 2020 Article ID 8931486 11 pageshttpsdoiorg10115520208931486

Emotion is a very complex psychological state producedby human beings in a specific environment which is oftenassociated with temperament temperament and motivation[1] People can feel their own emotional state at all times itprovides a guarantee for human survival and affects ourlearning decision-making and memory capabilities [2]Emotion is a personrsquos attitude and experience towardsobjective situations or things It is a physical and psycho-logical state produced by a personrsquos senses thoughts andbehaviors [3] Emotion occupies an important position inthe human society As an advanced function of the humanbrain it can ensure the adaptability of people in differentenvironments At the same time it can characterize humanpersonality characteristics and psychopathology [4] Gen-erally positive emotions can make people full of strengthand vitality and make people energetic so it is beneficial tophysical and mental health and the recovery of the bodyNeutral emotion is an important criterion for personalpsychological stability Negative emotions usually cause aperson to become depressed Being in this state for a longtime will affect peoplersquos working conditions and endangerphysical and mental health

erefore emotion recognition and monitoring has be-come a necessary way to solve human mental illness Humanemotion prediction also has important research significanceand application value in areas such as mental health evaluationFor example in medicine the relationship between emotionand stress and other diseases is studied by analyzing physio-logical signals such as EEG in different emotional states [5 6] Itis possible to find new ways to treat and recover similar mentalillnesses In education the distance teaching platform based onemotion recognition can become more humane by obtainingfeedback from students [7] In entertainment intelligentemotional sensing robots can bring more fun to life With thedeepening of research on emotion recognition its service areasfor humans will becomemore extensive At present in terms ofresearch materials emotion recognition can be divided intospeech-based [8 9] video-based [10 11] image-based [12ndash14]text-based [15 16] and physiological signal [17 18] andemotion recognition [19 20] combining multiple modal dataIn the recognition of classifiers they are mainly based onmachine learning [21ndash28] and based on deep learning [29 30]Machine learning algorithms have been successfully applied tothe recognition of various physiological signals [31ndash36] eapplication of deep learning algorithms is still being exploredfurther

is article is mainly devoted to the research of emotionrecognition based on EEG signals In sentiment analysisbased on EEG signals there are mainly two methods eyare linear analysis method and nonlinear analysis methodRepresentative studies are shown in Table 1

e abovementioned EEG-based emotion recognitionmethod does not consider the characteristics of differentrhythms in the EEG signal but processes the EEG uniformlyismethod ignores the different effects of different rhythmson emotion recognition Aiming at this problem this paperproposes an emotion recognition method based on thefusion of multirhythm results e contributions of thisresearch are summarized as follows

(1) In order to fully excavate the information charac-teristics of different rhythms in EEG signals thispaper extracts and classifies multiple rhythms ismethod can make full use of the information ofdifferent rhythms and has better pertinence

(2) Aiming at the problem of large feature dimensionspace and difficulty in integrating multiple rhythmsin emotion recognition in this study the D-S evi-dence combination theory was used to merge mul-tiple rhythm classification results to obtain the finalclassification results e result fusion method cannot only obtain more accurate results than a singlerhythm or simple integration of multiple rhythmsbut also reduce the dimension of the feature spaceand overcome the problem of how to integratemultiple rhythms

(3) is study used the FSVM classifier Due to theintroduction of the fuzzy membership mechanismthis classifier has better noise immunity than otherclassic classifiers is classifier is more suitable forapplications in noisy actual productionenvironments

2 Emotion Recognition Based on EEG Signals

21 Emotion Recognition Process Based on EEG Signale process of emotion recognition is essentially a process ofpattern recognition which is generally divided into threesteps ey are data collection and preprocessing featureextraction and model training and recognition Figure 1 is aflowchart of emotion recognition In the supervised machinelearning process it is first necessary to label the acquiredsample set divide it into different categories and divide thesample set into training set and test set Secondly datapreprocessing and feature extraction are required for the twosets Finally train the model by the training set e trainedmodel is for classification and decision-making In therecognition process the test set features are sent to thetrained model for sample prediction e output emotioncategory label is the recognition result thus completing thewhole process of emotion recognition

22 Introduction to EEG Signals According to differentclassification basis EEG signals can be divided into cate-gories as shown in Table 2

In the preprocessing of the received EEG signal noisereduction processing is mainly performed At the same timeit reduces the interference of non-brain wave signals such asskin electricity and muscle electricity en feature extrac-tion is performed on the data to obtain useful signals forsentiment analysis

23 Feature Extraction of EEG Signals In this study thewavelet transform was used to extract 4 rhythms in EEGelectrode signals namely θ rhythm α rhythm β rhythmand c rhythm Taking β as an example calculate the waveletpacket coefficients of the β wave decomposition node of the

2 Mathematical Problems in Engineering

Acquisition and preprocessing

Physiological signal

Video

Audio

Training data

Testing data

Feature extraction

Training sample

Testing sample

Training Model Emotion prediction

Figure 1 Flow chart of emotion recognition

Table 1 Representative research on emotion recognition based on EEG signals

Method Feature Representativeresearch

Recognitionrate ()

Linear analysis methods (Pearson correlationamplitude squared coherence autoregressivemodel cumulative energy algorithm time-frequency analysis etc)

EEG signal waveform characteristics (such asamplitude phase etc) rhythm wave averagepower power spectral density band energywavelet coefficient root mean square etc

References [6] 6042References [7] 6250

References [37] 8851

Nonlinear analysis methods (mutual information[38] correlation dimension LempelndashZiv (LZ)complexity recursive graph and entropyanalysis [39])

Entropy fractal dimension correlationdimension CO complexity LZ complexityHust index maximum Lyapunov index etc



Table 2 Classification of EEG signals

Classification basis Classification details

Frequency

(1) δ (01sim4Hz)(2) θ (4sim8Hz)(3) α (8sim13Hz)(4) β (13sim30Hz)(5) c (31sim100Hz)

Mathematical Problems in Engineering 3

EEG signal and obtain various statistical values of the EEGsignal through calculation ese original statistical valuesare used as original features According to the particularityand difference of the EEG signal the average energy of the βwave rhythm of the EEG signal in the time domain and thefrequency domain is extracted e characteristics of theextracted β waves are shown in Table 3

e calculation formula of some statistical values is asfollows

mean 1N

1113944

N

n1En

std

1N minus 1

1113944

N

n1En minus mean( 1113857

2

11139741113972

minRatio MinN

maxRatio MaxN

EneryMean 1N

1113944

N

n1E2n

(1)

where E represents the brain electrical signal data and Nrepresents the length of the brain electrical signal data

24 Learning and Classification of EEG Signals A supportvector machine (SVM) is one of the most common classi-fication methods in emotion recognition Considering thatthe classic SVM is susceptible to noise interference the EEGsignal collected in the real production environment usuallycontains noise interference In order to improve the clas-sification accuracy this paper uses the SVM with fuzzymembership

Let the training sample set be xi yi u(xi)1113864 1113865n

i1ix rep-

resents the feature vector of each sample yi represents two

different categories yi isin +1 minus1 u(xi) is the fuzzy mem-bership function u(xi) represents the membership degree ofthe ith sample and represents the reliability of the ith sample

xi belonging to the yi class 0lt u(xi)le 1 According to theprinciple of the SVM algorithm the training samples aremapped to the high-dimensional feature space and thefeature mapping function ϕ(middot) is used to obtain Rd⟶ RFe training sample is converted to ϕ(xi) yi u(xi)1113864 1113865 eclassification hyperplane is wlowast ϕ(xi) + b 0 where thekernel function represented by ϕ(middot) is K(xi xj)

ϕ(xi)Tϕ(xj)

min12

w2

+ C+

1113944

n

i1 | y+1

u+i ξi + C

minus1113944

n

i1 | yminus1

uminusi ξi

styi w middot ϕ xi( 1113857 + b1113858 1113859 minus 1 + ξi ge 0 i 1 2 n

ξ ge 0 i 1 2 n

⎧⎪⎪⎨

⎪⎪⎩

(2)

where Cminus and C+represent the penalty factors of positiveand negative samples respectively ξ is the relaxation factore optimal hyperplane is obtained by solving the objectivefunction by the Lagrangian multiplier method

f(x) sgn 1113944 αiyiK xi x( 1113857 + b1113872 1113873 (3)

According to the degree of influence of each sample onthe classification surface each sample point is given a dif-ferent degree of membership e purpose is to make thesample points with larger influence degree have a largerdegree of membership and the sample data with smallereffect will give a smaller degree of membership

25 7e D-S Evidence Combination 7eory Dempster firstdescribed the DempsterndashShafer evidence combination the-ory in his article [43] Later Shafer further developed andperfected the theory which formed the DempsterndashShaferevidence combination theory as it is now known eDempsterndashShafer evidence combination theory is also calledthe D-S evidence theory It expands the data fusion solutionand is widely used in multisource data fusion e D-Sevidence theory is based on the trust function of differentobservations and uses Dempsterrsquos evidence combination

Table 2 Continued


Gibbs classification

(1) Minor episode variability(2) Small waves

(3) High-amplitude slow wave(4) Low-speed slow wave

(5) Slow wave(6) 85sim120Hz step length is 05Hz

(7) Slow ground amplitude(8) Fast wave

(9) High-speed fast wave

Classification by EEG signal pattern

(1) αEEG(2) β EEG

(3) Flat EEG(4) Irregular EEG


rules to fuse them en a judgment is made on the resultobtained according to a certain type of rule and finally thefusion and final decision result is realized e principle isdescribed as follows [44]

Suppose a finite space Θ and let 2Θ be all the subsets inthe space Θ is also includes the empty setΘ itself For thesubset A define the function m 2Θ ⟶ (0 1) and satisfy

1113936AsubeΘ

m(A) 1

m(Φ) 0

⎧⎪⎨

⎪⎩(4)

Function m(A) is the basic confidence distributionfunction on Θ and m(A) is the precise trust level of thesubset A In this theory the basic confidence distributionfunction is assigned to A in a fixed form as its evidenceinformation However different people will give inconsis-tent confidence assignments to the same evidence because oftheir special experience and knowledge Maximize the use ofindependent and different sources of evidence to improvethe accuracy or confidence of the target event

m(D) 1k

1113944AcapBD

m1(A)m2(B) (5)

Assume

m(Φ) 0 k 1 minus 1113944AcapBΦ

m1(A)m2(B) (6)

3 The Proposed Emotion Recognition Method

Different rhythms in EEG data correspond to differentemotional states because emotion recognition based on asingle rhythm often has problems such as low recognitionrate and poor stability Using multiple EEG rhythms asfeature recognition will improve the recognition results andstability At present most emotion recognition based onmultiple EEG rhythms simply combine the feature recog-nition results extracted from these rhythms and there is nomore effective fusion strategy is makes the dimension ofthe feature space and the input dimension of the classifiertoo high making the accuracy and stability of the dis-crimination result poor In order to make full use of theadvantages of EEG data and improve decision-making re-sults this study applies the D-S theory to the decision-making levelemain idea of the proposedmethod is Firstextract the four characteristic rhythms in the EEG electrode

signal θ rhythm α rhythm β rhythm and c rhythm andextract each characteristic wave separately Secondly inputthe feature vector into the corresponding FSVM classifier forrecognition Finally the basic confidence distribution ofeach mode under each classifier is obtained and the D-Sevidence combination theory is used to fuse the classificationresults to obtain the final decision result e framework ofthe proposed method is shown in Figure 2

e steps of the proposed algorithm are as follows

Step 1 prepare the electrodes FC5-FC6 to be analyzedrespectively Due to the time-varying nonstationarycharacteristics of EEG preprocessing is essential beforewaveform extraction Among them there are mainlyframing and windowing Set the frame length to 512the frame shift to 256 and the window function to thehamming windowStep 2 extract the θ rhythm α rhythm β rhythm and c

rhythm of each electrode signal after preprocessingand use them as the EEG characteristic band efeature extraction is shown in Table 4Step 3 assign an FSVM classifier to identify eachrhythm Each FSVM can be regarded as independentevidence and its output value is transformed into thebasic confidence function of each emotionmodel underevidenceStep 4 after obtaining the basic allocation function ofeach FSVM classifier in step 3 perform fusionaccording to the formula (5)Step 5 the fusion result is judged according to the rulesm(Aw) max m(Ai)1113864 1113865 the function with the maxi-mum trust degree is selected as the target class

4 Experiment Analysis

41 Experimental Data and Parameter Settings e data setused in this study is a public data set provided by S Koelstraet al for analyzing the human emotional state e data setcontains audio and video 32-lead EEG data and 8-leadperipheral physiological signalse data set is divided into 4emotions namely high arousal and high valence (HAHV)low arousal and high valence (LAHV) low arousal and lowvalence (LALV) and high arousal and low valence (HALV)In the experiment 312 training samples were selected and144 test samples were selected Each type has 78 trainingsamples and 36 test samples is study used the data thatwere preprocessed by experimenters such as S Koelstra and

Table 3 Abbreviations and descriptions of signal statistical characteristics

Feature abbreviations DescriptionMean e average value of the β waveMedian Median of the β waveStd Standard deviation of the β waveMin Minimum value of the β waveMax Maximum value of β waveMin Ratio e ratio of the minimum number of β waves to the signal lengthMax Ratio e ratio of the maximum number of β waves to the signal lengthEnergy Mean Average energy of β wave


others after removing oculogram frequency reduction andfiltering from the original data e data sampling frequencyof each segment is reduced to 128Hz and the sampling timeis 63 seconds e first 3 seconds are the baseline durationand the next 60 seconds are the experimental data

e contrast classifiers used in this study are SVMGaussian mixture model (GMM) and BP neural network(BPNN) SVM the parameter c isin 2minus 5 251113864 1113865 and nuclearparameter c isin 2minus 5 251113864 1113865 e number of Gaussiancomponents is 6 e evaluation index is the recognitionrate

42 Experimental Program and Result Analysis is studydesigned experiments from three perspectives single mul-tirhythm as data input different classification result fusionstrategies and different classifiers e specific design plan isas follows

Scheme 1 In order to verify the influence of singlemultirhythm as data input on emotion recognition theexperiment compared the recognition rate of singlerhythm and multirhythm in different combinationsemultirhythm classification result of this experimentuses the fusion of the D-S evidence combination theoryand the classifier uses the FSVM e experimentalresults are shown in Table 5 and Figure 3Scheme 2 In order to verify the effectiveness of the D-Sfusion strategy used the two result fusion methods ofordinary linear combination and D-S evidence

combination are compared e experimental data usea combination of α+β+c three rhythms e experi-mental results are shown in Table 6Scheme 3 In order to verify the robustness of the FSVMclassifier the classic SVM GMM and BPNN are se-lected for the comparison classifier e experimentalresult data are a multirhythm fusion classification re-sult e experimental results are shown in Table 7 andFigure 4

From Table 5 and Figure 3 the following conclusions canbe drawn

(1) In a single rhythm the recognition rate of differentrhythms is different It shows that the contribution ofinformation carried by different rhythms is differentAmong them rhythms β andc have the better rec-ognition rate which shows that rhythms β and c cantruly reflect emotional changes When performingmultifeature recognition these two kinds of rhythmsshould be given priority

(2) e recognition rate of different combinations ofeach rhythm is higher than that of any single rhythmis shows that emotion recognition performancebased on multiple rhythm combinations is betterAmong the multirhythms of different combinationsthe D-S fusion recognition rate of the three rhythmsα+ β+ c is the highest e fusion recognition rate ofthe four rhythms θ + α+ β+ c is lower than the fu-sion recognition rate of the three rhythms α+ β+ c

Data preprocessing EEG signalEEG rhythm extraction

FSVM

FSVM

FSVM

FSVM

DS decision fusion

Featureextraction

Featureextraction

Featureextraction

Featureextraction

θ

α

β

γ

Decision output

Figure 2 Framework diagram of the proposed emotion recognition method

Table 4 Features extracted from each rhythm

Rhythm Feature detailsθ Time domain characteristics peak value mean value and standard deviation of time domain signal

Frequency domain characteristics power spectral density center of gravity frequency and frequency band energyNonlinear dynamic characteristics approximate entropy and sample entropy

αβc


Table 5 Comparison of the recognition rate between single rhythm and multirhythm recognition

Rhythm HALV LAHV LALV HALV Mean

Single rhythm

θ 05280 05426 04985 05446 05284α 05390 05566 05078 05335 05342β 05618 06309 04775 05821 05631c 05885 05321 05347 05538 05523

Multirhythm fusion

θ+ α 05311 05489 05026 05523 05337θ+ β 05602 06243 04724 05953 05631θ+ c 05925 05341 05387 05622 05569α+ β 05454 06243 04856 05754 05577α+ c 05565 05287 05295 05432 05395β+ c 05743 06265 05043 05754 05701

θ+ α+ β 06117 05876 05906 05350 05812θ++ α+ c 06243 05973 05667 05465 05837α+ β+ c 07088 05630 05894 06421 06258

θ+ α+β+ c 07008 05534 05878 06346 06192

θ α β γ0

01

02

03

04

05

06

07

08

Reco

gniti

on ra

te

HALVLAHVLALV

HALVMean

(a)

Figure 3 Continued


is shows that it is not that more rhythms are betterSome rhythms carry less useful information whichwill weaken the final decision result e recognitionrate of any combination of two rhythms is lower thanthe recognition rate of α+ β+ c three rhythmcombinations Among the three rhythm combina-tions the recognition rate of the α+ β+ c threerhythm combinations is significantly higher than theother three rhythm combinations is shows that inorder to obtain the optimal decision result not onlythe number of combined rhythms must be con-firmed but also the most representative rhythmmustbe selected

It can be concluded from Table 6 that the fusion methodbased on the D-S evidence combination has the highest

recognition rate e separate discrimination results of thethree rhythms are merged and the group with the largesttrust degree after fusion is taken as the target classe linearcombination experiment simply integrates the three rhythmfeatures into a set of feature vectors In the process of patternrecognition misjudgment may be caused due to the in-consistency between certain dimensional features From thecomparison of these two sets of experiments it can beconcluded that the D-S evidence combination theory canreduce the misjudgment caused by the inconsistency be-tween the features to a certain extent thereby improving therecognition rate

From Table 7 and Figure 4 the following conclusions canbe drawn the emotion recognition rate under the FSVMclassifier is the highest It is 302 higher than SVM 347higher than GMM and 133 higher than BPNN eoverall improvement is not large which shows that the use ofdifferent classifiers has little effect on the final recognitionrate Among different classifiers the recognition rates ofBPNN and FSVM are not much different is shows thatalthough the neural network-based classifier has highcomputational time complexity the final decision-makingeffect is ideal For some operations that do not consider timecost we can consider using a neural network-based classifierBy comparing the four emotion recognition results we

θ +

α

θ +

β

θ +

γ

β +

γ

α +

β

α +

γ

θ +

α +

β

θ +

α +

γ

α +

β +

γ

θ +

α +

β +

γ

0

01

02

03

04

05

06

07

08

Reco

gniti

on ra

te

HALVLAHVLALV

HALVMean

(b)

Figure 3 Comparison of recognition rates under different rhythm combinations

Table 6 e recognition rate of different result fusion strategies

Result fusion method HALV LAHV LALV HALV MeanLinear combination 06523 05532 05641 06330 06007D-S evidence combination theory 07088 05630 05894 06421 06258

Table 7 Comparison of recognition rates under differentclassifiers

Classifier HALV LAHV LALV HALV MeanSVM 06818 05432 05678 06346 06069GMM 06778 05528 05584 06302 06048BPNN 06972 05578 05708 06445 06176FSVM 07088 05630 05894 06421 06258


found that the recognition rates of HALV and HALV aregenerally higher Divided from the degree of arousal bothcategories belong to the range of high arousal is showsthat EEG emotional data with high arousal have a betterperformance in recognition

5 Conclusion

As the pressure of college students increases somenegative events occur frequently Emotion recognitionfor college students is particularly important andmeaningful In this context this article proposes an EEGsignal-based emotion recognition method for collegestudents First of all in the use of data sets this study usesmultirhythms as data input rough experimentalcomparison three rhythms were finally selected esecond step is to use the wavelet transform for featureextraction for each rhythm e third step is to use theFSVM classifier to classify the input feature data to obtainclassification results of different rhythms e fourth stepis to use the D-S evidence combination theory to fuse theclassification results of the three rhythms in order to getthe final decision result ere are 3 points of innovationin this research One is to use multiple rhythms as inpute second is to introduce the FSVM classifier withstrong noise immunity e third is to use the resultfusion strategy based on the D-S evidence theoryrough experimental comparison the emotion recog-nition method proposed in this article can effectivelyimprove the recognition rate which has a reference valueHowever this study also has some shortcomings forexample classifying different rhythms separately thismethod directly discards the relationship between dif-ferent rhythms and may reduce the final recognitioneffect Subsequently a classifier based on a collaborativelearning mechanism is used for classification andrecognition

Data Availability

e data used to support the findings of this study areavailable from the corresponding author upon request

Conflicts of Interest

e authors declare that they have no conflicts of interest

Acknowledgments

is work was supported by the National Natural ScienceFoundation of China (NSFC grant nos 51705021 U176426161702055 61972059 and 61773272) Key Laboratory ofSymbolic Computation and Knowledge Engineering Min-istry of Education Jilin University (93K172017K18)

References

[1] R Giner-Sorolla ldquoe past thirty years of emotion researchappraisal and beyondrdquo Cognition and Emotion vol 33 no 1pp 48ndash54 2019

[2] I Blanchette and A Richards ldquoe influence of affect onhigher level cognition a review of research on interpretationjudgement decision making and reasoningrdquo Cognition ampEmotion vol 24 no 4 pp 561ndash595 2010

[3] H Jazaieri A S Morrison P R Goldin and J J Gross ldquoerole of emotion and emotion regulation in social anxietydisorderrdquo Current Psychiatry Reports vol 17 no 1 p 5312014

[4] Z Rakovec-Felser ldquoe sensitiveness and fulfillment ofpsychological needs medical health care and studentsrdquoCollegium Antropologicum vol 39 no 3 pp 541ndash550 2015

[5] J A Healey and R W Picard ldquoDetecting stress during real-world driving tasks using physiological sensorsrdquo IEEETransactions on Intelligent Transportation Systems vol 6no 2 pp 156ndash166 2005

[6] H Sandler S Tamm U Fendel M Rose B F Klapp andR Bosel ldquoPositive emotional experience induced by

HALV LAHV LALV HALV Mean0

01

02

03

04

05

06

07

08

Reco

gniti

on ra

te

SVMGMM

BPNNFSVM

Figure 4 Comparison of recognition rates of different classifiers under multirhythm fusion


vibroacoustic stimulation using a bodymonochord in patientswith psychosomatic disorders is associated with an increasein EEG-eta and a decrease in EEG-alpha powerrdquo BrainTopography vol 29 no 4 pp 524ndash538 2016

[7] O K Akputu K P Seng Y Lee and L-M Ang ldquoEmotionrecognition using multiple kernel learning toward E-learningapplicationsrdquo ACM Transactions on Multimedia ComputingCommunications and Applications vol 14 no 1 2018

[8] S Nakagawa L Wang and S Ohtsuka ldquoSpeaker identifica-tion and verification by combining MFCC and phase infor-mationrdquo IEEE Transactions on Audio Speech and LanguageProcessing vol 20 no 4 pp 1085ndash1095 2012

[9] R Xia and Y Liu ldquoA multi-task learning framework foremotion recognition using 2D continuous spacerdquo IEEETransactions on Affective Computing vol 8 no 1 pp 3ndash142017

[10] H-W Yoo and S-B Cho ldquoVideo scene retrieval with in-teractive genetic algorithmrdquo Multimedia Tools and Applica-tions vol 34 no 3 pp 317ndash336 2007

[11] M Xu C Xu X He J S Jin S Luo and Y Rui ldquoHierarchicalaffective content analysis in arousal and valence dimensionsrdquoSignal Processing vol 93 no 8 pp 2140ndash2150 2013

[12] S-H Wang P Phillips Z-C Dong and Y-D Zhang ldquoIn-telligent facial emotion recognition based on stationarywavelet entropy and Jaya algorithmrdquo Neurocomputingvol 272 pp 668ndash676 2018

[13] Y Sun GWen and JWang ldquoWeighted spectral features basedon local Hu moments for speech emotion recognitionrdquo Bio-medical Signal Processing and Control vol 18 pp 80ndash90 2015

[14] A R Damasio T J Grabowski A Bechara et al ldquoSubcorticaland cortical brain activity during the feeling of self-generatedemotionsrdquo Nature Neuroscience vol 3 no 10 pp 1049ndash10562000

[15] C-H Wu Z-J Chuang and Y-C Lin ldquoEmotion recognitionfrom text using semantic labels and separable mixturemodelsrdquo ACM Transactions on Asian Language InformationProcessing (TALIP) vol 5 no 2 pp 165ndash182 2006

[16] D Zeng H Chen R Lusch and S-H Li ldquoSocial mediaanalytics and intelligencerdquo IEEE Intelligent Systems vol 25no 6 pp 13ndash16 2010

[17] W-L Zheng and B-L Lu ldquoInvestigating critical frequencybands and channels for EEG-based emotion recognition withdeep neural networksrdquo IEEE Transactions on AutonomousMental Development vol 7 no 3 pp 162ndash175 2015

[18] W-L Zheng J-Y Zhu and B-L Lu ldquoIdentifying stablepatterns over time for emotion recognition from EEGrdquo IEEETransactions on Affective Computing vol 10 no 3 pp 417ndash429 2019

[19] S Zhalehpour O Onder Z Akhtar and C E ErdemldquoBAUM-1 a spontaneous audio-visual face database of af-fective and mental statesrdquo IEEE Transactions on AffectiveComputing vol 8 no 3 pp 300ndash313 2017

[20] Y Wang L Guan and A N Venetsanopoulos ldquoKernel cross-modal factor analysis for information fusion with applicationto bimodal emotion recognitionrdquo IEEE Transactions onMultimedia vol 14 pp 597ndash607 2012

[21] P Qian Y Chen J-W Kuo et al ldquomDixon-based syntheticCT generation for PET attenuation correction on abdomenand pelvis jointly using transfer fuzzy clustering and activelearning-based classificationrdquo IEEE Transactions on MedicalImaging vol 39 no 4 pp 819ndash832 2020

[22] Y Jiang K Zhao K Xia et al ldquoA novel distributed multitaskfuzzy clustering algorithm for automatic MR brain imagesegmentationrdquo Journal of Medical Systems vol 43 no 5 2019

[23] P Qian C Xi M Xu et al ldquoSSC-EKE semi-supervisedclassification with extensive knowledge exploitationrdquo Infor-mation Sciences vol 422 pp 51ndash76 2018

[24] Y Jiang Z Deng F-L Chung et al ldquoRecognition of epilepticEEG signals using a novel multiview TSK fuzzy systemrdquo IEEETransactions on Fuzzy Systems vol 25 no 1 pp 3ndash20 2017

[25] P Qian J Zhou F Y Liang et al ldquoMulti-view maximumentropy clustering by jointly leveraging inter-view collabo-rations and intra-view-weighted attributesrdquo IEEE Accessvol 6 pp 28594ndash28610 2018

[26] Y Jiang D Wu Z Deng et al ldquoSeizure classification fromEEG Signals using transfer learning semi-supervised learningand TSK fuzzy systemrdquo IEEE Transactions on Neural Systemsand Rehabilitation Engineering vol 25 no 12 pp 2270ndash22842017

[27] P Qian Y Jiang Z Deng et al ldquoCluster prototypes and fuzzymemberships jointly leveraged cross-domain maximum en-tropy clusteringrdquo IEEE Transactions on Cybernetics vol 46no 1 pp 181ndash193 2016

[28] P Qian S Sun Y Jiang et al ldquoCross-domain soft-partitionclustering with diversity measure and knowledge referencerdquoPattern Recognition vol 50 pp 155ndash177 2016

[29] D Garg and G K Verma ldquoEmotion recognition in valence-arousal space frommulti-channel EEG data and wavelet baseddeep learning frameworkrdquo Procedia Computer Sciencevol 171 pp 857ndash867 2020

[30] S B Wankhade and D D Doye ldquoDeep learning of empiricalmean curve decomposition-wavelet decomposed EEG signalfor emotion recognitionrdquo International Journal of Uncer-tainty Fuzziness amp Knowledge-Based Systems vol 28 no 1pp 153ndash177 2020

[31] P C Petrantonakis and L J Hadjileontiadis ldquoEmotion rec-ognition from EEG using higher order crossingsrdquo IEEETransactions on Information Technology in Biomedicinevol 14 no 2 pp 186ndash197 2010

[32] Y Lin C H Wang T P Jung et al ldquoEEG-based emotionrecognition in music listeningrdquo IEEE Transactions on Bio-Medical Engineering vol 57 no 7 pp 1798ndash1806 2010

[33] Y Yang Q M J Wu W-L Zheng and B-L Lu ldquoEEG-basedemotion recognition using hierarchical network with sub-network nodesrdquo IEEE Transactions on Cognitive and Devel-opmental Systems vol 10 no 2 pp 408ndash419 2018

[34] Y Zhang Z Zhou W Pan et al ldquoEpilepsy signal recognitionusing online transfer TSK fuzzy classifier underlying classi-fication error and joint distribution consensus regulariza-tionrdquo IEEEACM Transactions on Computational Biology andBioinformatics 2020

[35] Y Jiang Y Zhang C Lin D Wu and C Lin ldquoEEG-baseddriver drowsiness estimation using an online multi-view andtransfer TSK fuzzy systemrdquo IEEE Transactions on IntelligentTransportation Systems 2020

[36] Y Zhang J Dong J Zhu and C Wu ldquoCommon and specialknowledge-driven TSK fuzzy system and its modeling andapplication for epileptic EEG signals recognitionrdquo IEEE Ac-cess vol 7 pp 127600ndash127614 2019

[37] A Greco G Valenza L Citi and E P Scilingo ldquoArousal andvalence recognition of affective sounds based on electroder-mal activityrdquo IEEE Sensors Journal vol 17 no 3 pp 716ndash7252017

[38] J Kim andEAndre ldquoEmotion recognition based onphysiologicalchanges inmusic listeningrdquo IEEETransactions on PatternAnalysisand Machine Intelligence vol 30 no 12 pp 2067ndash2083 2008

[39] M Wyczesany and T S Ligeza ldquoTowards a constructionistapproach to emotions verification of the three-dimensional


model of affect with EEG-independent component analysisrdquoExperimental Brain Research vol 233 no 3 pp 723ndash733 2015

[40] J Tao and T TanAffective Computing A review InternationalConference on Affective Computing and Intelligent Interactionvol 3784 pp 981ndash995 Springer Berlin Germany 2005

[41] K H Kim S W Bang and S R Kim ldquoEmotion recognitionsystem using short-term monitoring of physiological signalsrdquoMedical amp Biological Engineering amp Computing vol 42 no 3pp 419ndash427 2004

[42] S Paul A Banerjee and D N Tibarewala ldquoEmotional eyemovement analysis using electrooculography signalrdquo Inter-national Journal of Biomedical Engineering and Technologyvol 23 no 1 pp 59ndash70 2017

[43] A P Dempster ldquoUpper and lower probabilities induced by amultivalued mappingrdquo7e Annals of Mathematical Statisticsvol 38 no 2 pp 325ndash339 1967

[44] J Inglis ldquoA mathematical theory of evidencerdquo Technometricsvol 20 no 1 p 242 1976


Emotion is a very complex psychological state producedby human beings in a specific environment which is oftenassociated with temperament temperament and motivation[1] People can feel their own emotional state at all times itprovides a guarantee for human survival and affects ourlearning decision-making and memory capabilities [2]Emotion is a personrsquos attitude and experience towardsobjective situations or things It is a physical and psycho-logical state produced by a personrsquos senses thoughts andbehaviors [3] Emotion occupies an important position inthe human society As an advanced function of the humanbrain it can ensure the adaptability of people in differentenvironments At the same time it can characterize humanpersonality characteristics and psychopathology [4] Gen-erally positive emotions can make people full of strengthand vitality and make people energetic so it is beneficial tophysical and mental health and the recovery of the bodyNeutral emotion is an important criterion for personalpsychological stability Negative emotions usually cause aperson to become depressed Being in this state for a longtime will affect peoplersquos working conditions and endangerphysical and mental health

erefore emotion recognition and monitoring has be-come a necessary way to solve human mental illness Humanemotion prediction also has important research significanceand application value in areas such as mental health evaluationFor example in medicine the relationship between emotionand stress and other diseases is studied by analyzing physio-logical signals such as EEG in different emotional states [5 6] Itis possible to find new ways to treat and recover similar mentalillnesses In education the distance teaching platform based onemotion recognition can become more humane by obtainingfeedback from students [7] In entertainment intelligentemotional sensing robots can bring more fun to life With thedeepening of research on emotion recognition its service areasfor humans will becomemore extensive At present in terms ofresearch materials emotion recognition can be divided intospeech-based [8 9] video-based [10 11] image-based [12ndash14]text-based [15 16] and physiological signal [17 18] andemotion recognition [19 20] combining multiple modal dataIn the recognition of classifiers they are mainly based onmachine learning [21ndash28] and based on deep learning [29 30]Machine learning algorithms have been successfully applied tothe recognition of various physiological signals [31ndash36] eapplication of deep learning algorithms is still being exploredfurther

is article is mainly devoted to the research of emotionrecognition based on EEG signals In sentiment analysisbased on EEG signals there are mainly two methods eyare linear analysis method and nonlinear analysis methodRepresentative studies are shown in Table 1

e abovementioned EEG-based emotion recognitionmethod does not consider the characteristics of differentrhythms in the EEG signal but processes the EEG uniformlyismethod ignores the different effects of different rhythmson emotion recognition Aiming at this problem this paperproposes an emotion recognition method based on thefusion of multirhythm results e contributions of thisresearch are summarized as follows

(1) In order to fully excavate the information charac-teristics of different rhythms in EEG signals thispaper extracts and classifies multiple rhythms ismethod can make full use of the information ofdifferent rhythms and has better pertinence

(2) Aiming at the problem of large feature dimensionspace and difficulty in integrating multiple rhythmsin emotion recognition in this study the D-S evi-dence combination theory was used to merge mul-tiple rhythm classification results to obtain the finalclassification results e result fusion method cannot only obtain more accurate results than a singlerhythm or simple integration of multiple rhythmsbut also reduce the dimension of the feature spaceand overcome the problem of how to integratemultiple rhythms

(3) is study used the FSVM classifier Due to theintroduction of the fuzzy membership mechanismthis classifier has better noise immunity than otherclassic classifiers is classifier is more suitable forapplications in noisy actual productionenvironments

2 Emotion Recognition Based on EEG Signals

21 Emotion Recognition Process Based on EEG Signale process of emotion recognition is essentially a process ofpattern recognition which is generally divided into threesteps ey are data collection and preprocessing featureextraction and model training and recognition Figure 1 is aflowchart of emotion recognition In the supervised machinelearning process it is first necessary to label the acquiredsample set divide it into different categories and divide thesample set into training set and test set Secondly datapreprocessing and feature extraction are required for the twosets Finally train the model by the training set e trainedmodel is for classification and decision-making In therecognition process the test set features are sent to thetrained model for sample prediction e output emotioncategory label is the recognition result thus completing thewhole process of emotion recognition

22 Introduction to EEG Signals According to differentclassification basis EEG signals can be divided into cate-gories as shown in Table 2

In the preprocessing of the received EEG signal noisereduction processing is mainly performed At the same timeit reduces the interference of non-brain wave signals such asskin electricity and muscle electricity en feature extrac-tion is performed on the data to obtain useful signals forsentiment analysis

23 Feature Extraction of EEG Signals In this study thewavelet transform was used to extract 4 rhythms in EEGelectrode signals namely θ rhythm α rhythm β rhythmand c rhythm Taking β as an example calculate the waveletpacket coefficients of the β wave decomposition node of the




Video

Audio

Training data

Testing data

Feature extraction

Training sample

Testing sample





Recognitionrate ()











Frequency





mean 1N

1113944

N

n1En

std

1N minus 1

1113944

N


2

11139741113972

minRatio MinN

maxRatio MaxN

EneryMean 1N

1113944

N

n1E2n

(1)




i1ix rep-




ϕ(xi)Tϕ(xj)

min12

w2

+ C+

1113944

n

i1 | y+1

u+i ξi + C

minus1113944

n

i1 | yminus1

uminusi ξi


ξ ge 0 i 1 2 n

⎧⎪⎪⎨

⎪⎪⎩

(2)


f(x) sgn 1113944 αiyiK xi x( 1113857 + b1113872 1113873 (3)



Table 2 Continued









(1) αEEG(2) β EEG





1113936AsubeΘ

m(A) 1

m(Φ) 0

⎧⎪⎨

⎪⎩(4)


m(D) 1k

1113944AcapBD

m1(A)m2(B) (5)

Assume


m1(A)m2(B) (6)





















FSVM

FSVM

FSVM

FSVM

DS decision fusion

Featureextraction

Featureextraction

Featureextraction

Featureextraction

θ

α

β

γ

Decision output





αβc




Single rhythm

θ 05280 05426 04985 05446 05284α 05390 05566 05078 05335 05342β 05618 06309 04775 05821 05631c 05885 05321 05347 05538 05523

Multirhythm fusion

θ+ α 05311 05489 05026 05523 05337θ+ β 05602 06243 04724 05953 05631θ+ c 05925 05341 05387 05622 05569α+ β 05454 06243 04856 05754 05577α+ c 05565 05287 05295 05432 05395β+ c 05743 06265 05043 05754 05701

θ+ α+ β 06117 05876 05906 05350 05812θ++ α+ c 06243 05973 05667 05465 05837α+ β+ c 07088 05630 05894 06421 06258

θ+ α+β+ c 07008 05534 05878 06346 06192

θ α β γ0

01

02

03

04

05

06

07

08

Reco

gniti

on ra

te

HALVLAHVLALV

HALVMean

(a)

Figure 3 Continued






θ +

α

θ +

β

θ +

γ

β +

γ

α +

β

α +

γ

θ +

α +

β

θ +

α +

γ

α +

β +

γ

θ +

α +

β +

γ

0

01

02

03

04

05

06

07

08

Reco

gniti

on ra

te

HALVLAHVLALV

HALVMean

(b)








5 Conclusion


Data Availability




Acknowledgments


References








01

02

03

04

05

06

07

08

Reco

gniti

on ra

te

SVMGMM

BPNNFSVM















































Video

Audio

Training data

Testing data

Feature extraction

Training sample

Testing sample





Recognitionrate ()











Frequency





mean 1N

1113944

N

n1En

std

1N minus 1

1113944

N


2

11139741113972

minRatio MinN

maxRatio MaxN

EneryMean 1N

1113944

N

n1E2n

(1)




i1ix rep-




ϕ(xi)Tϕ(xj)

min12

w2

+ C+

1113944

n

i1 | y+1

u+i ξi + C

minus1113944

n

i1 | yminus1

uminusi ξi


ξ ge 0 i 1 2 n

⎧⎪⎪⎨

⎪⎪⎩

(2)


f(x) sgn 1113944 αiyiK xi x( 1113857 + b1113872 1113873 (3)



Table 2 Continued









(1) αEEG(2) β EEG





1113936AsubeΘ

m(A) 1

m(Φ) 0

⎧⎪⎨

⎪⎩(4)


m(D) 1k

1113944AcapBD

m1(A)m2(B) (5)

Assume


m1(A)m2(B) (6)





















FSVM

FSVM

FSVM

FSVM

DS decision fusion

Featureextraction

Featureextraction

Featureextraction

Featureextraction

θ

α

β

γ

Decision output





αβc




Single rhythm

θ 05280 05426 04985 05446 05284α 05390 05566 05078 05335 05342β 05618 06309 04775 05821 05631c 05885 05321 05347 05538 05523

Multirhythm fusion

θ+ α 05311 05489 05026 05523 05337θ+ β 05602 06243 04724 05953 05631θ+ c 05925 05341 05387 05622 05569α+ β 05454 06243 04856 05754 05577α+ c 05565 05287 05295 05432 05395β+ c 05743 06265 05043 05754 05701

θ+ α+ β 06117 05876 05906 05350 05812θ++ α+ c 06243 05973 05667 05465 05837α+ β+ c 07088 05630 05894 06421 06258

θ+ α+β+ c 07008 05534 05878 06346 06192

θ α β γ0

01

02

03

04

05

06

07

08

Reco

gniti

on ra

te

HALVLAHVLALV

HALVMean

(a)

Figure 3 Continued






θ +

α

θ +

β

θ +

γ

β +

γ

α +

β

α +

γ

θ +

α +

β

θ +

α +

γ

α +

β +

γ

θ +

α +

β +

γ

0

01

02

03

04

05

06

07

08

Reco

gniti

on ra

te

HALVLAHVLALV

HALVMean

(b)








5 Conclusion


Data Availability




Acknowledgments


References








01

02

03

04

05

06

07

08

Reco

gniti

on ra

te

SVMGMM

BPNNFSVM















































mean 1N

1113944

N

n1En

std

1N minus 1

1113944

N


2

11139741113972

minRatio MinN

maxRatio MaxN

EneryMean 1N

1113944

N

n1E2n

(1)




i1ix rep-




ϕ(xi)Tϕ(xj)

min12

w2

+ C+

1113944

n

i1 | y+1

u+i ξi + C

minus1113944

n

i1 | yminus1

uminusi ξi


ξ ge 0 i 1 2 n

⎧⎪⎪⎨

⎪⎪⎩

(2)


f(x) sgn 1113944 αiyiK xi x( 1113857 + b1113872 1113873 (3)



Table 2 Continued









(1) αEEG(2) β EEG





1113936AsubeΘ

m(A) 1

m(Φ) 0

⎧⎪⎨

⎪⎩(4)


m(D) 1k

1113944AcapBD

m1(A)m2(B) (5)

Assume


m1(A)m2(B) (6)





















FSVM

FSVM

FSVM

FSVM

DS decision fusion

Featureextraction

Featureextraction

Featureextraction

Featureextraction

θ

α

β

γ

Decision output





αβc




Single rhythm

θ 05280 05426 04985 05446 05284α 05390 05566 05078 05335 05342β 05618 06309 04775 05821 05631c 05885 05321 05347 05538 05523

Multirhythm fusion

θ+ α 05311 05489 05026 05523 05337θ+ β 05602 06243 04724 05953 05631θ+ c 05925 05341 05387 05622 05569α+ β 05454 06243 04856 05754 05577α+ c 05565 05287 05295 05432 05395β+ c 05743 06265 05043 05754 05701

θ+ α+ β 06117 05876 05906 05350 05812θ++ α+ c 06243 05973 05667 05465 05837α+ β+ c 07088 05630 05894 06421 06258

θ+ α+β+ c 07008 05534 05878 06346 06192

θ α β γ0

01

02

03

04

05

06

07

08

Reco

gniti

on ra

te

HALVLAHVLALV

HALVMean

(a)

Figure 3 Continued






θ +

α

θ +

β

θ +

γ

β +

γ

α +

β

α +

γ

θ +

α +

β

θ +

α +

γ

α +

β +

γ

θ +

α +

β +

γ

0

01

02

03

04

05

06

07

08

Reco

gniti

on ra

te

HALVLAHVLALV

HALVMean

(b)








5 Conclusion


Data Availability




Acknowledgments


References








01

02

03

04

05

06

07

08

Reco

gniti

on ra

te

SVMGMM

BPNNFSVM















































1113936AsubeΘ

m(A) 1

m(Φ) 0

⎧⎪⎨

⎪⎩(4)


m(D) 1k

1113944AcapBD

m1(A)m2(B) (5)

Assume


m1(A)m2(B) (6)





















FSVM

FSVM

FSVM

FSVM

DS decision fusion

Featureextraction

Featureextraction

Featureextraction

Featureextraction

θ

α

β

γ

Decision output





αβc




Single rhythm

θ 05280 05426 04985 05446 05284α 05390 05566 05078 05335 05342β 05618 06309 04775 05821 05631c 05885 05321 05347 05538 05523

Multirhythm fusion

θ+ α 05311 05489 05026 05523 05337θ+ β 05602 06243 04724 05953 05631θ+ c 05925 05341 05387 05622 05569α+ β 05454 06243 04856 05754 05577α+ c 05565 05287 05295 05432 05395β+ c 05743 06265 05043 05754 05701

θ+ α+ β 06117 05876 05906 05350 05812θ++ α+ c 06243 05973 05667 05465 05837α+ β+ c 07088 05630 05894 06421 06258

θ+ α+β+ c 07008 05534 05878 06346 06192

θ α β γ0

01

02

03

04

05

06

07

08

Reco

gniti

on ra

te

HALVLAHVLALV

HALVMean

(a)

Figure 3 Continued






θ +

α

θ +

β

θ +

γ

β +

γ

α +

β

α +

γ

θ +

α +

β

θ +

α +

γ

α +

β +

γ

θ +

α +

β +

γ

0

01

02

03

04

05

06

07

08

Reco

gniti

on ra

te

HALVLAHVLALV

HALVMean

(b)








5 Conclusion


Data Availability




Acknowledgments


References








01

02

03

04

05

06

07

08

Reco

gniti

on ra

te

SVMGMM

BPNNFSVM






















































FSVM

FSVM

FSVM

FSVM

DS decision fusion

Featureextraction

Featureextraction

Featureextraction

Featureextraction

θ

α

β

γ

Decision output





αβc




Single rhythm

θ 05280 05426 04985 05446 05284α 05390 05566 05078 05335 05342β 05618 06309 04775 05821 05631c 05885 05321 05347 05538 05523

Multirhythm fusion

θ+ α 05311 05489 05026 05523 05337θ+ β 05602 06243 04724 05953 05631θ+ c 05925 05341 05387 05622 05569α+ β 05454 06243 04856 05754 05577α+ c 05565 05287 05295 05432 05395β+ c 05743 06265 05043 05754 05701

θ+ α+ β 06117 05876 05906 05350 05812θ++ α+ c 06243 05973 05667 05465 05837α+ β+ c 07088 05630 05894 06421 06258

θ+ α+β+ c 07008 05534 05878 06346 06192

θ α β γ0

01

02

03

04

05

06

07

08

Reco

gniti

on ra

te

HALVLAHVLALV

HALVMean

(a)

Figure 3 Continued






θ +

α

θ +

β

θ +

γ

β +

γ

α +

β

α +

γ

θ +

α +

β

θ +

α +

γ

α +

β +

γ

θ +

α +

β +

γ

0

01

02

03

04

05

06

07

08

Reco

gniti

on ra

te

HALVLAHVLALV

HALVMean

(b)








5 Conclusion


Data Availability




Acknowledgments


References








01

02

03

04

05

06

07

08

Reco

gniti

on ra

te

SVMGMM

BPNNFSVM















































Single rhythm

θ 05280 05426 04985 05446 05284α 05390 05566 05078 05335 05342β 05618 06309 04775 05821 05631c 05885 05321 05347 05538 05523

Multirhythm fusion

θ+ α 05311 05489 05026 05523 05337θ+ β 05602 06243 04724 05953 05631θ+ c 05925 05341 05387 05622 05569α+ β 05454 06243 04856 05754 05577α+ c 05565 05287 05295 05432 05395β+ c 05743 06265 05043 05754 05701

θ+ α+ β 06117 05876 05906 05350 05812θ++ α+ c 06243 05973 05667 05465 05837α+ β+ c 07088 05630 05894 06421 06258

θ+ α+β+ c 07008 05534 05878 06346 06192

θ α β γ0

01

02

03

04

05

06

07

08

Reco

gniti

on ra

te

HALVLAHVLALV

HALVMean

(a)

Figure 3 Continued






θ +

α

θ +

β

θ +

γ

β +

γ

α +

β

α +

γ

θ +

α +

β

θ +

α +

γ

α +

β +

γ

θ +

α +

β +

γ

0

01

02

03

04

05

06

07

08

Reco

gniti

on ra

te

HALVLAHVLALV

HALVMean

(b)








5 Conclusion


Data Availability




Acknowledgments


References








01

02

03

04

05

06

07

08

Reco

gniti

on ra

te

SVMGMM

BPNNFSVM

















































θ +

α

θ +

β

θ +

γ

β +

γ

α +

β

α +

γ

θ +

α +

β

θ +

α +

γ

α +

β +

γ

θ +

α +

β +

γ

0

01

02

03

04

05

06

07

08

Reco

gniti

on ra

te

HALVLAHVLALV

HALVMean

(b)








5 Conclusion


Data Availability




Acknowledgments


References








01

02

03

04

05

06

07

08

Reco

gniti

on ra

te

SVMGMM

BPNNFSVM














































5 Conclusion


Data Availability




Acknowledgments


References








01

02

03

04

05

06

07

08

Reco

gniti

on ra

te

SVMGMM

BPNNFSVM













































emotionanalysisofcollegestudentsusingafuzzysupport...

Documents