Application and Comparison of ANN and SVM for Diagnostic ... · Application and Comparison of ANN...

Applied Mathematical Sciences, Vol. 10, 2016, no. 64, 3187 - 3199HIKARI Ltd, www.m-hikari.com

https://doi.org/10.12988/ams.2016.610258

Application and Comparison of ANN and SVM

for Diagnostic Classification for Cognitive Functioning

Yeliz Karaca1 and Sengul Hayta2

1on visiting Engineering School (DEIM)University of Tuscia, Viterbo, Rome, Italy

2Halic University, Faculty of EngineeringComputer Engineering Department, Istanbul, Turkey

Copyright c© 2016 Yeliz Karaca and Sengul Hayta. This article is distributed under the

Creative Commons Attribution License, which permits unrestricted use, distribution, and

reproduction in any medium, provided the original work is properly cited.

Abstract

The Wechsler Adult Intelligence Scale-Revised is a test designed andapplied comprehensively for classifying the cognitive skills of adults.Different intellectual profiles on a sample population of 414 individu-als, have been obtained to measure the relationship between the mentalfunctioning or psychopathological disorder of the individuals based onthe data through the Wechsler Adult Intelligence Scale-Revised test.The relevant data have been collected and considered as input neuronfor Artificial Neural Networks and Support Vector Machine Radial Ba-sis Function kernel Algorithms. Classification accuracy, time elapsedfor training procedure and accuracy rate have been obtained throughMulti Layer Perceptron. The training data is split as follows: 70% is thetraining data, 15% is the validation data and 15% is the test data. In ad-dition, in Support Vector Machine, time elapsed and accuracy rate havebeen evaluated based on 5-fold cross validation method. As a result,we will show the importance of Working Memory profile (Numbering,Assembly Ability) of The Wechsler Adult Intelligence Scale-Revised teston adults for the analysis of psychopathological disorder.

Keywords:The Wechsler Adult Intelligence Scale-Revised, Clinical Clas-sification, Artificial Neural Networks Algorithm, Support Vector Machine Al-gorithm

3188 Yeliz Karaca and Sengul Hayta

1 Introduction

The Wechsler Adult Intelligence Scale-Revised (WAIS-R) globally measuresand demonstrates adults’ capacity with respect to their voluntary movements,logical reasoning and coping with relevant factors at ease [9] [11] [14] [24] [29].WAIS-R is a test that measures intelligence which is regarded as a superiormental ability. The test is multidimensional and has multi determination fea-ture. The test is comprised of six verbal subtests and five performance sub-tests. Besides these subtests, the test is used for the calculation of verbal IQand functional IQ at universal terms [2] [5] [9, 10, 11] [14] [16] [21] [29]. Ref-erences cited [2] [4, 5] [10] [12, 13] [16] [18, 19] [21] [23] [30] present valuableinformation on the details and content concerning WAIS-R test [32].

WAIS-R is a test designed and applied for the research of adults cogni-tive skills comprehensively. Different intellectual profiles have been used tomeasure the correlation between General knowledge & Verbal fluency, Verbalconceptual, Attention & Concentration, Spatial visual motor coordination.These skills have been considered in order to measure the relationship be-tween the mental functioning of the patients and psychopathological data inthe WAIS-R test [32].In certain recent papers, these tests have also been takeninto consideration for manifesting the existence of a mental disorder [32].

Recent literature reveals that there have already been some attempts torelate the importance of WAIS-R test with the mental disorder. Yadav [26]and his team presented a new neuron model, with multiplicative neuron thatperforms multiplication operation rather than simple summation. Their sim-ulation results show that the proposed neuron model, used in a feed forwardneural network, performs better than existing multilayer networks (MLN).Cui[31] and his team studied how the artificial neural networks (ANN) can be usedto interpret student performance on cognitive diagnostic assessments (CDAs)and evaluate the performances of ANN using simulation results. The currentinvestigation focuses on the third component as researchers examine the useof ANN as a powerful nonparametric statistical method for diagnostic classifi-cation. Akande [17] and his team did a study on the importance of concretecompressive strength prediction in structure and building design. In theirpaper, they studied the performance of support vector machine (SVM). Theconventional method of determining the strength of concrete is complicatedand time consuming hence artificial neural network (ANN) is widely proposedin lie of this method. However, ANN is an unstable predictor due to thepresence of local minima in its optimization objective. It was found that SVMdisplayed a slightly better performance compared to ANN and is highly stable.Kalogirou [27] illustrated how artificial neural networks (ANN) and geneticalgorithms (GAs) played an important role in modelling and prediction of theperformance of various energy systems in buildings. Results presented prove

Application and Comparison of ANN and SVM ... 3189

the potential of artificial neural networks and genetic algorithms as designtools in many areas of energy applications in buildings. Gupta [1] focusedon developing empirical models for predicting surface roughness, tool wearand power required in turning operations. These response parameters weremainly dependent upon cutting velocity, feed and cutting time. Three com-peting data mining techniques, response surface methodology (RSM), artificialneural networks (ANN) and support vector regression (SVR), were applied indeveloping the empirical models. It was found that ANN and SVR modelswere much better than regression and RSM models for predicting the threeresponse parameters. Moraes [25] and his team studied document-level sen-timent classification to automate the task of classifying a textual review, asexpressing a positive or negative sentiment. On the benchmark data set ofMovies reviews, artificial neural network (ANN) outperformed support vectormachine (SVM) by a statistically significant difference, even on the contextof unbalanced data. Their results also confirmed some potential limitationsof both models like the computational cost of SVM at the running time andANN at the training time. Ren [15] showed the essential role of classificationof micro calcification clusters from mammograms in computer-aided diagno-sis for early detection of breast cancer, where (SVM) and (ANN) were twocommonly used techniques. A new strategy namely balanced learning withoptimized decision making was proposed to enable effective learning from im-balanced samples, which was further employed to evaluate the performance ofANN and SVM in this context. When the proposed learning strategy was ap-plied to individual classifiers, the results on the DDSM database demonstratedthat the performance from both ANN and SVM was considerably enhanced.Even though ANN outperformed SVM when balanced learning was absent, theperformance from the two classifiers became very similar when both balancedlearning and optimized decision making were employed. Quintana [22] and histeam studied mild cognitive impairment (MCI), a transitional state betweennormal aging and Alzheimer disease (AD). The aims of this study were to de-velop, train, and explore and develop the ability of ANN to differentiate MCIand AD, and to study the relevant variables in MCI and AD diagnosis. Theirresults indicated that ANN had an excellent capacity to discriminate MCIand AD patients from healthy controls. These findings provided evidence thatANN could be a useful tool for the analysis of neuropsychological profiles re-garding clinical syndromes. Karaca [32] applied WAIS-R data set on ArtificialNeural Networks with 13 healthy individuals and 401 patients. The algorithmsin the study were Feed Forward Back Propagation and Cascade Forward BackPropagation. The aim of the team’s experimental studies was to measure thediagnostic classification success using these algorithms. The best performancein classification belonged to Feed Forward Back Propagation algorithm.

Classification of the disorder among algorithms was compared with clas-


sification performances. While doing the classification with Artificial NeuralNetworks Multi Layer Perceptron (ANN - MLP) algorithm and Support VectorMachine Radial Basis Function (SVM - RBF) kernel, two different data setswere used. The first data set consists of data obtained from profiles like Ver-bal Comprehension, Perceptual Reasoning, Working Memory and ProcessingSpeed. The second data set is comprised of data which has been obtained afterhaving removed the Working Memory profile (Numbering, Assembly Ability)from the first set of data. The study underlines the necessity of accuracy whileapplying the Working Memory of WAIS-R test on the patients. This studyincludes data of WAIS-R test administered on both females and males. It isaimed to contribute to literature since it intends to identify the classificationof individuals as healthy and unhealthy through machine medium.

2 Material and Methods

The materials under investigation are the data sets taken through a WAIS-Rtest on a population of 414 healthy and unhealthy individuals. The analyticalmethods applied on these data sets are derived from ANN - MLP and SVM -RBF kernel algorithms.

2.1 Data sets

For this study, the test has been applied on 414 individuals, constituting bothfemales(4 of them are healthy, 196 of them are unhealthy individuals) andmales (9 of them are healthy, 205 of them are unhealthy individuals) aged be-tween 14-87. The relevant data for the study has been collected over a period often years from the Unit of Psychiatry of Department of Clinique Neuroscience,Surgery of Psycho diagnostic of the Federico II University of Naples. The toolused to evaluate the intellectual profile of patients is WAIS-R scale (WechslerAdult Intelligence Scale Revised). WAIS-R has been used on grounds of relia-bility and frequency of use as tools across the world. Furthermore, it ensuresthe estimation of various mental processes as to the cognitive functioning ofadults. As for the education levels of the individuals, it can be stated thatthere are different groups that can be defined as follows: 3rd grade, 5th grade(Elementary School),8th grade (Middle School), 12th grade (High school) andUniversity degree [32].

This study has parameters that include problems regarding the mentalfunctioning and psychopathology (affective, mental and behavioral scopes) ofthe patients. For Data Set 1, 21 parameters are the following: School Edu-cation, Gender, Age, Similarity, Logical Deduction, Vocabulary, Information,Memory, Comparison, Verbal Information Volume (VIV), Quantity Informa-tion Verbal (QIV), Finiteness or Closure, Ordering Ability, 3 Dimensional


Modeling, Quantity Information Performance (QIP), Arithmetic, AssemblyAbility, Quantity Information Total (QIT), Performance Information Verbal(VIP), Performance Information Verbal (VIT), Full Scale Intelligence Quo-tient (DM). Data Set 2 includes the application of 19 parameters are SchoolEducation, Gender, Age,Similarity, Logical Deduction, Vocabulary, Informa-tion, Memory, Comparison, VIV, QIV, Finiteness or Closure, Ordering Abil-ity, 3 Dimensional Modeling, VIP, QIP, QIT, VIT, DM. There are two classes,Healthy/Unhealthy, on the output layer of the application. The reason forcarrying out the evaluation with two different data sets was the necessity ofapplying the profiles taken from WAIS-R test thoroughly. The reason forcarrying out the evaluation with two different data sets was the necessity ofapplying the profiles taken from WAIS-R test thoroughly. If such profiles arenot applied thoroughly, the diagnosis of the disorder cannot be accurate. Thisis of great significance both for physicians while diagnosing the disorder andfor those who perform classification in machine learning algorithm.

2.2 Methodology

General structure of the ANN - MLP and SVM - RBF kernel algorithms be-longing to the study and the significant data in the diagnosis of disorder appliedon the input layer of these methods are explained.

2.2.1 Artificial Neural Network

The structure of a basic artificial neuron is a lot simpler than that of a biologi-cal neuron. There are mainly data obtained from the external environment orother neurons, namely inputs, weights, addition function, activation functionand outputs in an artificial neuron. Data received from outside is bound tothe neuron by means of the weights. The weights determine the impact ofthe relevant input. The total function is calculated by the net input. Thenet input is a product of the relevant weights. The activation function calcu-lates the net output during the operation which also yields the neuron output[3] [6] [20] [31].

O = f (W.X + b)W = w1, w2, w3, ...., wn

X = x1, x2, x3, ...., xn

(1)

As it can be seen from Equation 1, Equation 2, Equation 3 Output (O) isthe product of the X inputs and the weights (W) applied onto the X inputs.

net =n∑

i=1

wixi + b (2)


O = f (net) (3)

The denotation above can be used [32].

2.2.2 Multi Layer Perceptron

Multi layer perceptron (MLP) is a feed-forward ANN system where more thanone layer is used between the input and output layers. In these medium layers,named as hidden layer, there are procedural elements whose knots are notdirectly bound to the input and output layers. The value on the input doesnot change until the weights reach the appropriate point. The outputs arecalculated with the desired responses, and if required error is stated. Errordenotation is used for changing the weights from the hidden procedure elementsto the output unit.Instead, the impact of each unit on the output unit errorsshould be known. This is done by taking the sum of the error denotations ofthe output units bound to hidden unit for each errored unit [6] [20] [28].Based on the equations 2,3,4,5; x input x0 =+1 is added into the input layerand hence expanded.

W = sigmoid(wTh x) =

1

1 + exp[−∑d

j=1 whjxj + wh0

] , h = 1, . . . , H (4)

yi outputs make up the second learning layer which gets the input neurons inthe hidden layer.

yi = vTi z =H∑

h=1

vihzh + vi0 (5)

There always exists an additional value unit shown as z0 in the hiddenlayer, its value is always +1 and it shows the weights vi0 going towards theoutputs.

2.3 Support Vector Machine

Support Vector Machine (SVM) is a modern classifier. SVM is capable of doinggood generalizations structuring linear classification boundaries in multidimen-sional space by means kernel [7] [8] [28]. SVM design is split by hyperplane[7] that signifies the decision boundaries named “support vectors”. The classprediction is performed based on these decision boundaries.

In SVM, a maximum level of accuracy is attained in the estimation ofclass prediction concerning a new data set which is formed through the use ofoptimum decision boundary from the training data. Afterwards, its accuracy


rate has been analyzed by obtaining class accuracy [7]. The following can beindicated for SVM with two classes:

wTxt + w0 ≥ +1ifrt = +1 (6)

wTxt + w0 ≤ −1ifrt = −1 (7)

The application is conducted with a vector which is able to be split lin-early as 1 or -1 with class representation based on the Equations 6 and 7 [8].The hyperplanes of the samples are not found merely adjacent to the line.For a better generalization they are supposed to endure within a certain dis-tance. The distance of the nearest samples on both sides of the boundary isthe margin; and this margin should be as high as possible for an optimumgeneralization [8].

3 Results and Discussion

A record of 414 individuals, both healthy and unhealthy subjects, is includedin our study. Data received according to WAIS-R has been applied as inputfor the ANN - MLP and SVM - RBF kernel algorithms.There are two out-put neurons on the output layer representing the unhealthy and the healthyindividuals.

Machine learning tool was MATLAB [33] used for the ANN - MLP andSVM - RBF kernel model. During our experimental studies, the same param-eters were used for Data Set 1 and Data Set 2 in MLP algorithm. Sigmoidfunction as the transfer function on the hidden layer and linear transfer func-tion for the output layer were made through a network structure that is a 10layered advanced version network. The hidden layer neuron number for DataSet 1 and Data Set 2 in MLP architecture has been selected as 10 . MLPlearning parameters are as follows: validation set proportion has been selectedas 0.20, and learning rate as 0.15. The training procedure is maximum 100iterations, error threshold value is 0.01, and momentum is set as 1.00e-08.Based on Figure 1 a) according to Data Set 1 in WAIS-R, as a result of thetraining procedure that lasted 42 seconds for 10 iterations, the performancevalue has been obtained as 0.00118.For the validation procedure set as Epoch8 the best validation result has been obtained at Epoch 2. Based on Figure 1b) according to Data Set 2 in WAIS-R, as a result of the training procedurethat lasted 2 minutes and 17 seconds for 46 iterations , the performance valuehas been obtained as 1.94e-09. The best validation result has been yielded atEpoch 46.

It has been observed that when Numbering and Assembly Ability param-eters are removed from the data set, training procedure takes a longer time.


Figure 1: a) MLP Classifier Performance for Data Set 1 b) MLP ClassifierPerformance for Data Set 2.

While the time elapsed is 42 seconds for Data Set 1, it is 2 minutes and 17seconds for Data Set 2. When Mean Squared Error is analyzed, the follow-ing can be stated: through the removing of Numbering and Assembly abilityparameters from Data Set 2, it has been seen that a better result in classifi-cation accuracy has been obtained compared to Data Set 1 for all the threeprocedures, namely training, validation and test.

Figure 2: a) Confusion matrix of SVM RBF kernel algorithm based on DataSet 1 5-fold cross validation procedure b) Confusion matrix of SVM RBF kernelalgorithm based on Data Set 2 5-fold cross validation procedure.

Based on Figure 2 a) related to Data Set 1 5-fold cross validation procedure,as a result of training procedure of SVM RBF kernel algorithm lasted 5 seconds,it yielded an accuracy performance for classification amounting to 96.9%, errorrate was obtained as 3.1% . 400 of 414 instances have been classified accurately.

Based on Figure 2 b) related to Data Set 2 5-fold cross validation procedure,as a result of training procedure of SVM RBF kernel algorithm lasted 7 seconds,it yielded an accuracy performance for classification amounting to 96.1%, errorrate was obtained as 3.9% of 414 instances have been classified accurately. In


conclusion, the accuracy rate performance in classification obtained by usingSVM algorithm RBF kernel proved to be better in Data Set 1 compared to thatin Data Set 2. In the class representation for WAIS-R data set belonging toindividuals through ANN and SVM, the relationship between the Numberingand Assembly Ability parameters are presented in Figure 3.

Figure 3: Thin-plate spline interpolant graph through the class representationof WAIS-R Numbering and Assembly Ability attributes.

Although a better classification performance has been obtained in DataSet 2 in MLP, the time elapsed for the training procedure is 1 minute and 31seconds longer than that of Data Set 1. For the SVM RBF kernel training pro-cedure, there has been found out to be a 0.8% better classification performancein Data Set 1 compared to Data Set 2.

4 Conclusions

Administered on 414 individuals for the identification of the patients prob-lems regarding their mental functioning, issues related to affect, cognition andbehavior WAIS-R test has provided data for this study. Data received fromthe individuals has been applied on MLP and SVM (RBF kernel) which areamong the machine learning methods. Based on the test, the classification ofthe individuals as healthy and unhealthy has been performed. Two differentdata sets have been taken as reference while obtaining the classification per-formance. Data Set 1 includes all the parameters obtained from WAIS-R test.Data Set 2 is comprised of units after having the Numbering and AssemblyAbility parameters removed from Data Set 1. These two Data Sets have beenapplied on ANN - MLP and SVM - RBF kernel Algorithms. The study haspresented the significance of applying WAIS-R parameters in a through wayfor the diagnosis of the individuals disorders. Hence, it can be said that thisstudy can be cited as a reliable system which physicians can benefit from inthe field of medicine. This study provides contribution to literature since itstresses the importance of disorder identification in machine medium regardingthe WAIS-R parameters. The contribution can be related to the comparisonof accuracy rate derived from the disorder classification and the required time


for the training procedures with respect to ANN-MLP and SVM-RBF kernelalgorithms. Besides, as the number of data in our database increases, this ratecould prove be a good reference not only for the algorithms in this study butalso for other methods available in the scope of engineering.

Acknowledgements. Dr. Yeliz Karaca would like to thank Prof. DianaGalletta, her team at the Unit of Psychiatry of Department of Clinique Neu-roscience, Surgery of Psychodiagnostic of the Federico II University of Naplesfor their input in the study for sharing their data on mental functions. Theauthors would also like to express their appreciation to the patients in thestudy.

References

[1] A.K. Gupta, Predictive Modelling of Turning Operations using ResponseSurface Methodology, Artificial Neural Networks and Support Vector Re-gression, International Journal of Production Research, 48 (2010), no. 3,763-778. https://doi.org/10.1080/00207540802452132

[2] A.S. Kaufman, J.E. Mclean, C.R. Reynolds, Sex, Race, Residence,Region and Education Differences on the 11 WAIS-R Subtests, Journalof Clinical Psychology, 44 (1988), no. 2, 231-248.https://doi.org/10.1002/1097-4679(198803)44:2<231::aid-jclp2270440224>3.0.co;2-j

[3] A. Robert, J. Diederich, A.B. Tickle, Survey and Critique of Techniquesfor Extracting Rules from Trained Artificial Neural Networks, Knowledge-based Systems, 8 (1995), no. 6, 373-389.https://doi.org/10.1016/0950-7051(96)81920-4

[4] A.S. Kaufman, L. Elizabeth, Assessing Adolescent and Adult Intelligence,(3rd ed.), Wiley, Hoboken, New Jersey, 2006.

[5] B. Hayslip, P.E. Panek, Adult Development and Aging, Second Edition,Harper Collins College Publishers, 1993.

[6] B. Karlık, Soft Computing Methods in Bioinformatics: A ComprehensiveReview, Mathematical and Computational Applications, 18 (2013), no. 3,176-197. https://doi.org/10.3390/mca18030176

[7] C. Cortes, V. Vapnik, Support-Vector Networks, Machine Learning, 20(1995), 273-297. https://doi.org/10.1007/bf00994018

[8] C. Cortes, V. Vapnik, Statistical Learning Theory, Wiley, New York, 1998.


[9] C.R. Brand, The Factor: General Intelligence and its Implications, Wiley,Chichester, UK, 1996.

[10] D. Galletta, C. Aurino, G. Sica, A. Amodio, C. Elce, F. Micanti, Self-Injurious Behavior: Self Identity, Impulsiveness and Self-Injury in Pa-tients with Borderline Personality Disorders and Bulimia, Journal of Psy-chiatry, 18 (2015), 262. https://doi.org/10.4172/psychiatry.1000262

[11] D. Wechsler, Wechsler Adult Intelligence Scale - Revised, TX: The Psy-chological Corporation, New York, San Antonio, 1955.

[12] D. Weschler, WAIS-R Manual: Wechsler Adult Intelligence Scale Revised,The Psychological Corporatio, New York, 1981.

[13] J.A. Athanasou, Patterns of Performance on the Verbal and PerformanceSubtests of the Wechsler Adult Intelligence Scale- Revised: SomeAustralian data, Journal of Clinical Psychology, 49 (1993), 102-108.https://doi.org/10.1002/1097-4679(199301)49:1<102::aid-jclp2270490114>3.0.co;2-3

[14] J.M. Arceneaux, G.M. Chearamine, C.W. Smith, Gender Differences inWAIS-R Age Corrected Scaled Scores, Perceptual and Motor Skills, 83(1996), 1211-1215. https://doi.org/10.2466/pms.1996.83.3f.1211

[15] J. Ren, ANN vs. SVM: Which one Performs Better in Classification ofMCCs in Mammogram Imaging, Knowledge-Based Systems, 26 (2012),144-153. https://doi.org/10.1016/j.knosys.2011.07.016

[16] K. Hattori, R. Lynn, Male-Female Differences on Japanese WAIS-R, Per-sonal Individual Differences, 23 (1997), no. 3, 531-533.https://doi.org/10.1016/s0191-8869(97)80021-9

[17] K.O. Akande, T.O. Owolabi, S. Twaha, S.O. Olatunji, Performance Com-parison of SVM and ANN in Predicting Compressive Strength of Concrete,IOSR J. Comput. Eng, 16 (2014), no. 5, 88-94.https://doi.org/10.9790/0661-16518894

[18] L. Atkinson, Some Tables for Statistically Based Interpretation of WAIS-R Factor Scores, Psychological Assessment, 3 (1991), 288-291.https://doi.org/10.1037//1040-3590.3.2.288

[19] L. Atkinson, Concurrent use of Wechsler Memory Scale Revised and theWAIS,- R*, British Journal of Clinical Psychology, 30 (1991), 87-90.https://doi.org/10.1111/j.2044-8260.1991.tb00923.x


[20] M.C. Bishop, Pattern Recognition and Machine Learning, Springer-VerlagNew York, 2006.

[21] M.H. Daniel, Intelligence Testing-Status and Trends, American Psychol-ogist, 52 (1997), no. 10, 1038-1045.https://doi.org/10.1037/0003-066x.52.10.1038

[22] M. Quintana, J. Guardia, G. Sanchez-Benavides, M. Aguilar, J.L. Molin-uevo et al., Using Artificial Neural Networks in Clinical Neuropsychology:High Performance in Mild Cognitive Impairment and Alzheimer’s disease,Journal of Clinical and Experimental Neuropsychology, 34 (2012), no. 2,195-208. https://doi.org/10.1080/13803395.2011.630651

[23] R.L. Adams, J. Smigielski, R.L. Jenkins, Development of Satz - MogelShort form of the WAIS-R, Journal of Consulting and Clinical Psycholog-ical, 52 (1994), 908. https://doi.org/10.1037//0022-006x.52.5.908

[24] R.M. Kaplan, D.P. Saccuzzo, Psychological Testing: Principles, Appli-cations, and Issues, (8th ed.), Cengage Learning Belmont, Wadsworth,Canada, 2010.

[25] R. Moraes, F.V. Joao, P.G.N. Wilso, Document-level Sentiment Classifica-tion: An Empirical Comparison Between SVM and ANN, Expert Systemswith Applications, 40 (2013), no. 2, 621-633.https://doi.org/10.1016/j.eswa.2012.07.059

[26] R.N. Yadav, P.K. Kalra, J. John, On the use of Multiplicative Neuronin Feedforward Neural Networks, International Journal of Modelling andSimulation, 26 (2006), no. 4, 331-336.https://doi.org/10.2316/journal.205.2006.4.205-4384

[27] S.A. Kalogirou, Artificial Neural Networks and Genetic Algorithms inEnergy Applications in Buildings, Advances in Building Energy Research,3 (2009), 83-119. https://doi.org/10.3763/aber.2009.0304

[28] V. Vapnik, Statistical Learning Theory, Wiley, New York, 1998.

[29] Wechsler Adult Intelligence Scale–Revised, list of tests Available fromthe CPS Testing Library, Center for Psychological Studies at University.Retrieved 2009-03-31.

[30] W. David, The Measurement of Adult Intelligence, Baltimore (MD):Williams and Witkins, 1939. https://doi.org/10.1037/10020-000


[31] Y. Cui, M. Gierl, Q. Guo, Statistical Classification for Cognitive Diag-nostic Assessment: An Artificial Neural Network Approach, EducationalPsychology, 36 (2015), no. 6, 1065-1082.https://doi.org/10.1080/01443410.2015.1062078

[32] Y. Karaca, Case Study on Artificial Neural Networks and Applications,Applied Mathematical Sciences, 10 (2016), no. 45, 2225-2237.https://doi.org/10.12988/ams.2016.65174

[33] http://www.mathworks.com (2015a Release).

Received: October 30, 2016; Published: December 3, 2016

Application and Comparison of ANN and SVM for Diagnostic ... · Application and Comparison of ANN...

Documents

Transcript of Application and Comparison of ANN and SVM for Diagnostic ... · Application and Comparison of ANN...