1999年度日本オペレーションズ・リサーチ学会
秋季研究発表会
1−F−6
6匠N匠陀A』Ⅶ＄AV¶⑳N◎『N匠M陀A』M匠WW⑳陀陀M⑳防匠』≠N阿隠匿田口6VⅢN6M⑳MS監 囲R旺6匿障臣陀F⑳RMÅN6臣
Ku Ruhana Ku−Mahamud AzuratizaAbu Bakar
NoritaMd．Norwawi
＄6帆⑳◎Ⅲ⑳晋ヨm陶『maせ毒⑳mV◎6陶m◎8⑳gy O Mm転v即＄菰朋M臨『aMaⅥay馴a
HousIngindustryhas been averyJmPOrtantSeCtOrinthe MaIays］an eCOnOmic
growth．The［ackofacomprehensivenationa［houslng POlicyhas resultedina
hightyspeculativehouslngPnCemarket．LiteraturereviewsreveaItheneedsto haveasystematicmodeltoovercometheprobIemofpredictingthehouseprlCe
in Ma［aysla．Buyers can use the modelto seeif a house has been pr［Ced
COrreCtly and on the other hand deve10PerS Can Checkifa profit Orloss has
incu汀ed．
The study concentrates on the used of neuraInetwork（NN）pa止icuIarly the
muIt‖ayer perceptron（MLP）too］in predicting the price of terrace housesin
SeLangor，uSlnglNSPEN as the majOr reference．The abiIity to dealwith
nonparametric variabtes，Which are foundin the houslngindustry，is a maJOr
advantage ofNN modelling．NNs are easyto usefbrfast pattern recognition Withoutbuilding a system ofsimuItaneous equations and ableto automatically
determine／capturepossib［elatemfunctjonsexistinginhistorica（data，Withoutthe intevention of subjective hypothesis．It also has s［gnificant advantages over COnVentionalrule or frame based，eXPe沌 SyStem aPPrO声Chesin some applications since NN do not requlre knowledge to be formalized．Other
techniquessuchasregressionandheuristicscouldnotgiveapreciseprediction
−138−
© 日本オペレーションズ・リサーチ学会. 無断複写・複製・転載を禁ず.
inthepncemovementsbecausepr［CemOVementSfbrecastinglSaneXamPleof
thereaJworldsysteminthattheychangeovertime．
Houslngdatafrom1994tol997whichwerereleasedbylNSPEN，WereuSedfor
Predictingthepnces．VariabJesljnvoIvedindetermjnJngtheprJCearetheyearthe
datawascoIIected，Iandandbuilduparea．typeoflandownership，tyPeandage Ofthehouse，distancefromtown，theenvironmentand bu‖dingqua［ity．Data Were Pre−Prbcessed，rePreSented and rescated．before being separatedinto
trainlngandtestsets．Thepre−PrOCeSSlngaCtivityhastakencareofthemisslng Variables，defdu（t va（ues and unre［iab（e andinconsistent data fieldsin the houslngdatabase．Variab］es＜thathavebeenrepresentedbybinarypatternOorl arethetypeofterrace houses，tyPeOf［and ownership，aSWeIIas aTea and bujJdingquaJities．VariabJeswhichwererescaJedareJandandbujJdupar甲，age
bfthe house，distancefromtown andthe houseprlCe．Rescaling processwas
Performedsoastorepresenttheva［uesintherangebetweenOandl．
The best pnce modet consists ofonelnPutlayerwith nine nodes，One OutPut layerwithonenodeandonehiddentayerwithfive nodes．Thepredictiveand
g9neratisationabi［ities・Ofthemode］weretested・ResultsobtainedfromMLPare
COmParab［e to the ones obtained from，regreSSion，Which shows th白t MLP P？S甲SS COnSiderabJe potentiaJas an aIternative to regression fbr prediction
PurPOSeSaSWe asusefulandeffectivein modeI［ingandanalyslng・realestate
markets．
Thisstudyhasreveatedthatfunherworkcanbedonetoenhancetheobtained
Pr−Ce．mOdeL・lssueslike data pre■−PrOCeSS−ng teChnique，initiatlweight fbr Simulation，PrlCe mOVementandfactors such a岳theJeconomicgrowth，incdnle
per capita and hedonic characteristic can be considered in the process of
PrOduc．ng the pr．ce modeI・Furtherworkcan aIso bedoneon other・tyPeSOf houses such as bung早Iows and flats．This studytherefore has opened new avenuesforfutureresearch．
−139−
© 日本オペレーションズ・リサーチ学会. 無断複写・複製・転載を禁ず.
