Appled Mathematcal Scences, Vol. 7, 2013, no. 41, 2047-2053 HIKARI Ltd, www.m-hkar.com Modelng Mult Layer Feed-forward Neural Network Model on the Influence of Hypertenson and Dabetes Melltus on Famly Hstory of Heart Attack n Male Patents 1 Norzan Mohamed, 2 Wan Muhamad Amr W Ahmad, 3 Nor Azlda Aleng and 4 Maza Hura Ahmad 1,2,3 Jabatan Matematk, Fakult Sans dan Teknolog, Malaysa Unverst Malaysa Terengganu (UMT) 21030 Kuala Terengganu, Terengganu, Malaysa azlda_aleng@umt.edu.my (Nor Azlda Aleng) 4 Jabatan Matematk, Fakult Sans, Unverst Teknolog Malaysa 81030 UTM Skuda, Johor, Malaysa. Copyrght 2013 Norzan Mohamed et al. Ths s an open access artcle dstrbuted under the Creatve Commons Attrbuton Lcense, whch permts unrestrcted use, dstrbuton, and reproducton n any medum, provded the orgnal work s properly cted. Abstract Accordng to prevous research fndngs, both hypertenson and dabetes melltus ncreases the chance of havng a stroke, heart attack, heart falure and kdney dsease. Patents wth dabetes and hypertenson have a hgher ncdence of coronary artery dsease compared to the patents wth dabetes or hypertenson alone. The am of ths study was to evaluate the nfluence of hypertenson and dabetes melltus on famly hstory of heart attack n male patents from the vew of Mult Layer Feed Forward (MLFF) neural network model. The nput varables of MLFF neural network model are selected based on the sgnfcant varables of logstc regresson (LR) model whch were, ncdent (new) coronary heart dsease (INCCHD), total cholesterol (CHOLTOT) and heght (cm). These three varables also were sgnfcantly contrbuted to MLFF neural network model.
2048 Norzan Mohamed et al. Keywords: Mult layer feed forward neural network model, multple logstc regresson, hypertenson and heart attack INTRODUCTION Nowadays, most of people lfestyles have grown very hectc due to the busy lfestyles. Consequently, the tme for rest and relaxaton become very lmted. Excess workload from offce has caused psychologcal tensons. Ths stuaton n our body can be translated as tensons rase the adrenalne n the blood and ths causes our blood pressure rse. When the blood pressure rse, the pressure force blood aganst the walls of arteres. Ths ncreases the rsk of stroke, aneurysm, heart falure, heart attack and kdney damage [1]. Hgh blood pressure and dabetes melltus share certan physologcal trats and they usually tend to occur together. Patent wth dabetes melltus wll ncreases the amount of flud n the body, when flud n the body ncrease, t wll lead to correspondng ncreases n blood pressure [2]. Besdes that, dabetes also ncreased arteral stffness and t decreases the ablty of the blood vessels to stretch. As a result the averages of blood pressure ncrease. Accordng to the prevous research, researcher found that common bologcal trats of the two dseases are lkely to occur together smply because they share a common set of rsk factors [3]. Multlayer feed-forward (MLFF) neural network was appled to medcal data [4, 5]. Norzan Mohamed et al. [4] develop a multple lnear regresson (MLR) and two multlayer feed-forward (MLFF) neural network models of body mass ndex (BMI). Model 1 was developed based on all the ndependent varables whch are systolc blood pressure (SBP), dastolc blood pressure (DBP), age, serum cholesterol (CHOLES), gender, pulse, heght, weght, arm, leg, wast and wrst. Model 2 was developed based on four sgnfcant ndependent varables of multple lnear regresson. The result shown that Model 2 was more accurate than Model 1. In 2012, they also appled the MLFF to lymphoma cancer data [6]. Recently, Nor Azlda Aleng et al. [7] appled MLFF neural network model on tuberculoss patents. The combnatons of sgnfcant varables from multple lnear regresson (MLR) model were used as nput varables of MLFF neural network model. The MLR of tuberculoss showed that gender, marred status, dseases gental and descendant were sgnfcant. Hence, four varables were used to develop the best (MLFF) neural network model of tuberculoss. In ths paper, patent wth hgh blood pressure and dabetes melltus have been analyzed n order to dentfy the factors that contrbute to the heart attack. The obectve of the current study s to study from the vew of MLFF neural network model the most sgnfcant varables that nfluences of hypertenson and dabete
Modelng mult layer feed-forward neural network model 2049 melltus on famly hstory of heart attack n male. The nput varables of MLFF neural network model are selected based on the sgnfcant varables of logstc regresson ( LR) model. Ths study focus on male patents wth hgh blood pressure and type II dabetes melltus. A total of 92 elgble patents were selected and explanaton of the varables as presented n Table 1. They were dagnosed to have dabetes melltus and hgh blood pressure (140mmHg or hgher) based on WHO crtera. A fastng plasma glucose level >126 mg/dl (7.0 mmol/l) or a casual plasma glucose >200 mg/dl (11.1 mmol/l) meets the threshold for the dagnoss of dabetes. The selected varables are: Famly hstory of heart attack (Fhha), total cholesterol (Choltot), ncdent (new) coronary heart dsease (Incchd) and Heght (cm). Table 1. Explanaton of the Varables Varables Code Explanaton of the varables Categorcal FHHA Y Famly hstory of heart attack 0 = No 1 = Yes INCCHD 1 Incdent (new) coronary heart dsease durng 6 years of follow-up CHOLTOT 2 Total cholesterol(mg/dl) HEIGHT 3 Heght (cm) 0 = No 1 = Yes METHOD Multlayer Feed-forward Neural Network (MLFF) Multlayer Feed-forward Neural Network (MLFF) conssts of an nput layer, one or several hdden layers and an output layer. The neurons n the feed-forward neural network are generally grouped nto layers. Sgnals flow n one drecton from the nput layer to the next, but not wthn the same layer [8]. An essental factor of successes of the neural networks depends on the tranng network. Among the several learnng algorthms avalable, back-propagaton has been the most popular and most wdely mplemented [9]. Bascally, the BP tranng algorthm wth three-layer feed-forward archtecture means that, the network has an nput layer, one hdden layer and an output layer. In ths research the output node s fxed at one snce there s only one ndependent varable. Thus, for the feed-forward network wth N nput nodes, H hdden nodes and one output node, the values Yˆ are gven by: H Yˆ = g w h + w0 (1) = 1 where w s an output weght from hdden node to output node, w 0 s the bas for output node, and g s an actvaton functon. The values of the hdden nodes h,
2050 Norzan Mohamed et al. = 1, K, H are gven by: N h = k v 0, = 1, K, H (2) = 1 Here, v s the nput weght from nput node to hdden node, v 0 s the bas for hdden node, are the ndependent varables where = 1, K, N and k s an actvaton functon. The archtecture of the multlayer feed-forward neural network model s llustrated n Fgure 1. Fg. 1: The archtecture of the multlayer feed-forward neural network model wth N nput nodes, H hdden nodes and one output node. RESULT AND DISCUSSION Logstc regressons are used extensvely n the medcal research and also help to make medcal decsons. Ths paper extended the dea of LR to MLFF neural network to examne the sgnfcant varables that nfluencng the hypertenson and dabetes melltus on famly hstory of heart attack. In 2012, Wan Muhammad Amr et al. [10] found that three ndependent varables whch ncdent of coronary heart dsease ( 1 ), total of cholesterol ( 2 ) and heght ( 3 ) sgnfcantly nfluence n the logstc regresson model. Hence these three ndependent varables ( 1, 2 and 3 ) are used as the nput varables of MLFF model. Norzan Mohamed et al. [4] used the sgnfcant varables from MLR model as nput nodes of MLFF neural network model. Instead of sgnfcant varables from MLR model, the sgnfcant varables from LR model are used n ths current study. The output node n ths study s one node snce we have one dependent varable whch s the famly hstory of heart attack (FHHA). To fnd the approprate number of hdden nodes and best combnaton of nput varables, the model selecton
Modelng mult layer feed-forward neural network model 2051 strateges whch proposed by Norzan Mohamed [11] s followed. The best number of hdden nodes s one node. Usng dfference combnaton of sgnfcant varables, result shown that the best combnaton s three nput varables as llustrated n Table 2. The archtecture of current study s composed of three nput varables, one hdden node and one output node as presented n Fgure 2. Thus, MLFF neural network model wth three nput varables, one hdden node and one output node, the values Yˆ are gven by: ( w h w ) Yˆ = g 1 + 0 = 1 (3) where w 1 s an output weght from hdden node to output node, w 0 s the bas for output node, and g s lnear functon. The values of the hdden nodes h, = 1 are gven by: 3 h = k v 0 = 1 (4) = 1 Here, v s the nput weght from nput node to hdden node, v 0 s the bas for hdden node, are the ndependent varables where = 1,2, 3 and k s a sgmod functon. Equatons 3 and 4 can also be represented as follows: Y ˆ = w + w h (5) h 0 1 1 [ [ ( ) ] 1 = 1+ exp v 0 1 1 2 2 3 3, = 1 (6) The archtecture of MLFF neural network model s llustrated n Fgure 2. Fg. 2: The archtecture of MLFF neural network model wth three nput varables, one hdden node and one output node.
2052 Norzan Mohamed et al. Table 2: The results of MSE tranng and MSE testng. Input Varables MSE Tranng MSE Testng 1, 2, 3 0.1894 0.1383 1, 2 0.1944 0.1452 1 0.1985 0.1985 1, 2, and 3 are INCCHD, CHOLTOT and HEIGHT respectvely CONCLUSION The purpose of the current study s to examne from the vew of MLFF neural network model the most sgnfcant varable that contrbute to famly hstory of heart attack n male patents. The combnatons of sgnfcant varables from logstc regresson (LR) model are used as nput varable of MLFF neural network model. Three varables were ncdent of coronary heart dsease ( 1 ), total of cholesterol ( 2 ) and heght ( 3 ) contrbute sgnfcantly to LR model. Usng the sgnfcant varables of LR model, the performance of MLFF neural network model for dfference combnatons are tested. The performance of MLFF was evaluated MSE of testng/out-sample. The combnaton of three nput varables outperformance other combnatons. Hence, these three varables also sgnfcantly contrbute to MLFF neural network model. REFERENCES [1] M.G. Scott, J. B. Ivor, L. B. Gregory, C. Alan, H. E. Robert, V. H. Barbara, W. Mtch, C. S. Sdney and R. S. James, Dabetes and cardovascular dsease: a statement for healthcare professonals from the Amercan Heart Assocaton. Crculaton. Amercan Heart Assocaton, Inc., 100: 1134-1146, 1999. [2] P. Aschner, Current concept of dabetes melltus. Int. Opthalmol Cln, 38: 1-10, 1998. [3] J.R. Sowers, Treatment of hypertenson n patents wth dabetes. Arch. Intern. Med. 164(17), 1, 850 857, 2004. [4] Norzan Mohamed, Wan Muhamad Amr W. Ahmad, Nor Azlda Aleng and Mazah Hura Ahmad, Assessng the Effcency of Multlayer Feed-Forward Neural Network Model: Applcaton to Body Mass Index Data. World Applcaton Sc. Journal, 15 (5): 677-682, 2011.
Modelng mult layer feed-forward neural network model 2053 [5] Nor Azlda Aleng, Norzan Mohamed, Wan Muhamad Amr W Ahmad and Ny Ny Nang, A New Strategy to Analyze Medcal Data usng Combnaton of M-Estmator and Multlayer Feed-Forward Neural Network Model. European Journal of Scentfc Research, 73(1), 79-85, 2012. [6] Norzan Mohamed, Nor Azlda Aleng, Wan Muhamad Amr W Ahmad and Mazah Hura Ahmad, Multlayer Feed-Forward Neural Network Approach to Lymphoma Cancer Data. Int. J. Contemp. Math. Scences, 7(35), 1749 1756, 2012. [7] Nor Azlda Aleng, Norzan Mohamed and Wan Muhamad Amr W Ahmad, Modellng of tuberculoss patents usng multlayer feed forward neural network. Appled Mathematcal Scences, 6 (50): 2481 2488, 2012. [8] T.D. Pham and. Lu, Neural Networks for Identfcaton, Predcton and Control, Great Brtan, 1995. [9] G.A. Darbellay and M. Slama, Forecastng the short-term demand for electrcty. Do neural networks stand a better chance? Internatonal Journal of Forecastng, 16:71-83, 2000. [10] Wan Muhamad Amr W Ahmad, Norzan Mohamed, Nor Azlda Aleng, and Zalla Al, Influence of Hypertenson and Dabetes Melltus on Famly Hstory of Heart Attack n Male Patents. Appled Mathematcal Scences, 6(66): 3259 3266, 2012. [11] Norzan Mohamed, Parametrc and artfcal ntellgence based methods for forecastng short term electrcty load demand. Unpublshed PhD thess, 2011. Receved: December 21, 2012