.
Development of an Expert System for Diagnosisand appropriate Medical Prescription of HeartDisease Using SVMand RBF
Shaikh AbdulHannan
1
,
1
Lecturer, Departmentof Computer Science,Vivekanand College,Aurangabad,Maharashtra, India,
hannan_7us@yahoo.com
V. D. Bhagile
2
,
2
Lecturer, Departmentof Computer Science,Sayali College,Aurangabad,Maharashtra, India.
v_shinde68@yahoo.co.in
R. R. Manza
3
,
3
Lecturer, Dept. of Comp. Science and I.T.Dr. B.A.M.University,Aurangabad,Maharashtra
manzaramesh@gmail.com
R. J. Ramteke
4
4
Reader, Dept. of Computer Science, North MaharashtraUniversity,Jalgaon, Maharashtra,
rakeshramteke@yahoo.co.in
Abstract---
Human disease diagnosis is a complicated process and requireshigh level of expertise. Any attempt of developing a expert systemdealing with human disease diagnosis has to overcome variousdifficulties. This paper describes a project work aiming todevelop a expert system for diagnosing of heart disease usingneural network technique. Now a days neural network are beingused successfully in an increasing number of application areas.This project includes the detailed information about patient andpreprocessing was done. The Support Vector Machine (SVM)and Radial Basis Function (RBF) have been applied over the datafor the experiment. This research project focuses on the researchand development of a web-based clinical tool designed to improvethe quality of the exchange of health information between healthcare professionals and patients. Practitioners can also use thisweb-based tool to corroborate diagnosis. The proposed system isexperimented on various scenarios in order to evaluate it’sperformance. In all the cases, proposed system exhibitssatisfactory results.
Keywords
–Neural Network, Support Vector Machine (SVM), RBF (Radial Basis Function), Symptoms, Medicine.
I.INTRODUCTION
Expert systems(ES) are artificial intelligence-base computer programs that have received a great deal of attention duringyears. These programs have been used to solve an impressivearray of problems in a variety of fields. The part of the expertsystem that stores the knowledge is called the knowledge base.The part that holds the specifics of the to-be-solved problem iscall the global database. The part that applies the knowledgeto the problem is called the inference engine. Expert systemstypically have friendly user interface to enable inexperiencedusers to specify problems for the system to solve and tounderstand the system’s conclusion.Although, there are many computer based diagnosissystems are developed for medicine. [1-3], however thenumber of expert system for human heart diagnosis domainsare still very few [4].
The Need for Expert Systems
Expert systems are necessitated by the limitations associatedwith conventionalhuman decision-making processes,including:1. Human expertise is very scarce.2. Humans get tired from physical or mental workload.3. Humans forget crucial details of a problem.4. Humans are inconsistent in their day-to-day decisions.5. Humans have limited working memory.6. Humans are unable to comprehend large amounts of dataquickly.7. Humans are unable to retain large amounts of data inmemory.8. Humans are slow in recalling information stored inmemory.9. Humans are subject to deliberate or inadvertent bias in their actions.10. Humans can deliberately avoid decision responsibilities.11. Humans lie, hide, and die.The proposed methodologyuses neural network for classifier. The performance of proposed methodology wasevaluated with two different neural network techniques.Moreover, we compared our result with Support Vector Machine and Radial Basis Function with original medicines provided by the doctor. We obtain 97% accuracy from theexperiments made on the data set containing 300samples.This classification is the highest so for with our data.
The paper is organizedas following, in Section 2, a brief overviewon previous related works and in section 3, introduction of Support Vector Machine and Radial Basis Functionisdescribed. Section 4, the proposed methodology and preparingData for underlying neural network. Section 5,Experimental
1.
Shaikh Abdul Hannan
-Workingas lecturer in Vivekanand College,Aurangabad and persuing Ph.D. from North Maharashtra University, Jalgaon(hannan_7us@yahoo.com)2.
V. D. Bhagile
-Working as Principal in Sayali College, Aurangabad.(v_shinde68@yahoo.com)3.
R. R. Manza
-Working as Lecturer in Department of Computer Scienceand Information Technology, Dr. B. A. M. University, Aurangabad.(manzaramesh@gmail.com)4.
R. J. Ramteke
-Working as a reader in Department of Computer Science, North Maharashtra University,Jalgaon. (rakeshramteke@yahoo.co.in)
IJCSIS) International Journal of Computer Science and Information Security,Vol. 8, No. 5, August 2010245http://sites.google.com/site/ijcsis/ISSN 1947-5500
.
analysis and how the coding is done with patients as well asmedicine data is described. Section 6,Discussion and resultof first five patients medicine given by the expert system andis compared with the original medicine. Finally, we concludedthis paper in Section 7.
II. BACKGROUND
A.Related Works
Up to now, various classification algorithms have beenemployed on Turkoglu’s valvular heart disease data set andhigh classification accuracies have been reported in the lastdecade [5-10]. Turkoglu’s valvular heart disease data set wasobtained from Firat Medical Center. A detailed description for the data set will be given in the next section.The valvular heart disease data set was firstly utilized in [5]where Turkoglu et al. fulfilled an expert diagnosis systemwhich uses backpropagation artificial neural networks (BP-ANN) classifier. The performance evaluation of the realizedsystem was evaluated by classification accuracy and thecorrect classification rate was about 94% for normal subjectsand 95.9% for abnormal subjects. Later, Turkoglu et al.suggested an intelligent system for detection of heart valvedisease based on waveletpacket neural networks (WPNN) [6].The reported correct classification rate was about 94% for abnormal and normal subjects. Recently, Comak et al.investigated the use of least-square support vector machines(LS-SVM) classifier for improving the performance of theTurkoglu’s proposal [7]. They intended to realize acomparative study. Classification rates of the examinedclassifiers were evaluated by ROC curves based on the termsof sensitivity and specificity. The application results showedthat according to the ROC curves, the LS-SVM classifier performance was comparable with ANN, but the training timeof LS-SVM is shorter than that of the ANN and it can alwaysconverge the same solution while ANN cannot. According tothese results, LS-SVM’s training time is about 13 timesshorter than ANN’s training time. This is an importantdifference. Because, LS-SVMs are trained only depending onsupport vectors, not by whole training data set. In addition,LS-SVM can overcome the overfitting much successfully thanANN.More recently, Uguz et al. performed a biomedical system based on Hidden Markov Model for clinical diagnosis andrecognition of heart valve disorders [8]. The fulfilledmethodology was also used the database of Turkoglu et al. Inthe presented study, continuous HMM (CHMM) classifier system was used. Single Gaussian model was preferred todetermine emission probability. The methodology wascomposed of two stages. At the first stage, the initial values of average and standard deviation were calculated by separatingobservation symbols into equal segments according to thestate number and using observation symbols appropriate toeach segment. At the second stage, the initial values of average and standard deviation were calculated by separatingobservation symbols into the clusters (FCM or K-meansalgorithms) that have equal number of states and usingobservation symbols appropriate to the separated clusters. Theimplementations of the experimental studies were carried outon three different classification systems such as CHMM,FCM–K-means/CHMM and ANN. These experimental resultswere obtained for specificity and sensitivity rates 92% and94% for CHMM, 92% and 97.26% for FCM–K-means/CHMM), respectively. Finally, Sengur et al.investigated the use of principal component analysis (PCA),artificial immune system (AIS) and fuzzy
k
-NN to determinethe normal and abnormal heart valves from the Doppler heartsounds [9]. For reducing the complexity, PCA was used. In theclassification stage, AIS and fuzzy
k
-NN were used. Toevaluate the performance of the methodology, a comparativestudy was realized by using a data set containing 215 samples.The validation of the method was measured by using thesensitivity and specificity parameters; 95.9% sensitivity and96% specificity rate was obtained. Sengur et al. alsoinvestigated the use of Linear Discriminant Analysis (LDA)and Adaptive neuro-fuzzy inference system (ANFIS) for clinical diagnosis and recognition of heart valve disorders[10]. The validation of the method is measured by using thesensitivity and specificity parameters. 95.9% sensitivity and94% specificity rate was obtained.
III. ORGANIZATION TO EXPERT SYSTEM
To diagnose the heart failure cause different popular methods used are MRI, Doppler and Expert System. MRI can provide clear three dimensional images of the heart. Doppler technique has gained much more interest since Satomura firstdemonstrated the application of the Doppler effect to themeasurement of blood velocity in 1959[11]. However thefactor such as calcified disease or obesity often result in adiagnostically unsatisfactory. Doppler techniques assessmentand therefore, it is sometimes necessary to assess thespectrogram of the Doppler Shift signals to elucidate m thedegree of the disease [12]. Many studies have beenimplemented the classify Doppler signals in the patternrecognition field [13-14].Expert system is a intelligent program which olds theaccumulated knowledge of one or more domain experts.There are many types of expert system currently exist.MYCIN expert system is used in medical field for diagnosis of blood disorders. DESIGN ADVISOR is another expertsystem used in processor chip design to give advice todesigner about component placement, minimizing chip sizeetc. PUFF expert system is also used in medical system for diagnosis of respiratory condition of patient. PROSPECTOR expert system is used by geologists to identify sites for drillingor mining. DENDRAL expert system is used to identify thestructure of chemical compounds. LITHIAN expert systemgives advice to archaeologists to examiner stone tools. Expertsystem having three main components knowledge base,inference engine and user interact. Knowledge base is thecollection of facts and rules which describe all the knowledgeabout problem domain. The inference engine is used tochoose the appropriate facts and rules to apply during user query. Where as user interface takes the user query in areadable form and passes it to the inference engine. It thendisplays the result tothe user. Based on much more useful but it has some limitation like limited domain, no currentupdation, no system self learning, no common sense, expertneeded to setup and maintain. But even though it is used
IJCSIS) International Journal of Computer Science and Information Security,Vol. 8, No. 5, August 2010246http://sites.google.com/site/ijcsis/ISSN 1947-5500
.
world wide because they are not always available, can be usedanytime anywhere, human experts are not 100% reliable or consistent. Expert may not good for explanation of decisionsand cost effective. While using the expert system some legaland ethical issues we need to follow to set theresponsibility.[15]
A.Support Vector Machine :
Support Vector Machines(SVMs) are a state of the art patternrecognition techniques whose foundation stem from statisticallearning theory. However, the scope of SVMs goes beyond pattern recognition because they can also handle two morelearning problems i.e. regression estimation and densityestimation. An SVM is a general algorithm based onguaranteed risk bounds of statistical learning theory i.e. the socalled structural risk minimization principle. It is a learningmachine capable of implementing s set of functions thatapproximate best the supervisor’s response with an expectedrisk bounded by the sum of the empirical risk and Vapnik – Chevonenkis (VC) confidence. Recent advances in statistics,generalization theory, computational learning theory, machinelearning and complexity have provided new guidelines anddeep insights into the general characteristics and nature of themodel building/learning/fitting process [16]. Some researchershave pointed out that statistical and machine learning modelsare not all that different conceptually [17,18]. Many of thenew computational and machine learning methods generalizethe idea of parameter estimation in statistics. Among thesenew methods, Support Vector Machines have attracted mostinterest in the last few years.Support vector machine (SVM) is a novel learningmachine introduced first by Vapnik [19]. It is based on theStructural Risk Minimization principle from computationallearning theory. Hearst et al. [20] positioned the SVMalgorithm at the intersection of learning theory and practice:‘‘it contains a large class of neural nets, radial basis function(RBF) nets, and polynomial classifiers as special cases. Yet itis simple enough to be analyzed mathematically, because itcan be shown to correspond to a linear method in a highdimensional feature space nonlinearly related to input space.’’In this sense, support vector machines can be a good candidatefor combining the strengths of more theory-driven and easy to be analyzed conventional statistical methods and more data-driven, distribution free and robust machine learning methods.In the last few years, there have been substantialdevelopments in different aspects of support vector machine.These aspects include theoretical understanding, algorithmicstrategies for implementation and reallife applications. S VMhas yielded excellent generalization performance on a widerange of problems including bioinformatics [21,22,23], textcategorization [24], image detection [25], etc. Theseapplication domains typically have involved high-dimensionalinput space, and the good performance is also related to thefact that SVM’s learning ability can be independent of thedimensionality of the feature space.The SVM approach has been applied in several financialapplications recently, mainly in the area of time series prediction and classification [26,27]. A recent study closelyrelated to our work investigated the use of the SVM approachto select bankruptcy predictors. They reported that SVM wascompetitive and outperformed other classifiers (includingneural networks and linear discriminant classifier) in terms of generalization performance [28]. In this study, we areinterested in evaluating the performance of the SVM approachin the domain of heart disease in comparison with that of Radial Basis Function in neural networks.Let us define labeled training examples [xi, yi], an inputvector
ni
R x
a class value
l i y
i
,.....,1,1,1
Forthelinearly separable case, the decision rules defined by anoptimal hyperplane separating the binary decision classes isgiven as the following equation in terms of the support vectorswhere Y is the outcome, yi is the class value of the trainingexample xi, and . represents the inner product. The vector x=(x1,x2,.,xn) corresponds to an input and the vectors xi,i=1,.,N, are the support vectors. In Eq. (1), b and
i
are parameters that determine the hyperplane.For the non-linearly separable case, a high-dimensionalversion of Eq. (1) is given as follows:The function K(x,xi) is defined as the kernel function for generating the inner products to construct machines withdifferent types of non-linear decision surfaces in the inputspace.
B.Radial Basis Function (RBF)
Radial basis function (RBF) networks wereintroduced into the neural network literature by Broomheadand Lowe [29]. The RBF network is similar to a general feed-forward neural network trained using the back-propagationscheme. It has three layers of neurons, namely input, hiddenand output. However it uses only one hidden layer, eachneuron in which operates as the Gaussian transfer function, asagainst the sigmoid function [30]Mathematically, the output
y
of an RBF network corresponding to input
x
is computed by the equation;
01
||||)(
wc xw x f y
inii
(Eq.A.1)
Where
w
i
is the connection weight between the
i
th hiddenneuron and output neuron;
w
0
the bias.
φ||x
-c
i
||
indicates aRBF which is normally Gaussian havingfollowing expression;
)2||||exp(||||
122
nii
iii
c xc x
(Eq. A.2)
Wherec
i
are centers of the receptive field; and σ
i
the widths of the Gaussian function which indicates the selectivity of aneuron.
IJCSIS) International Journal of Computer Science and Information Security,Vol. 8, No. 5, August 2010247http://sites.google.com/site/ijcsis/ISSN 1947-5500
Search History:
Result 00 of 00
00 results for result for
p.
Development of an Expert System for Diagnosis and appropriate Medical Prescription of Heart Disease Using SVM and RBF
Human disease diagnosis is a complicated process and requires high level of expertise. Any attempt of developing a expert system dealing with human disease diagnosis has to overcome various difficulties. This paper describes a project work…
(More)
1.9K
Reads
3
Readcasts
8
Embed Views
Get Scribd Mobile
To get Scribd mobile enter your number and we'll send you a link to the Scribd app for iPhone & Android.We've sent a link to the Scribd app. If you didn't receive it, try again.
We'll never share your phone number.
Recommended
More From This User
Notes
Load more
