I.J.IntelligentSystemsandApplications,2012,9,22-28PublishedOnlineAugust2012inMECS()DOI:10.5815/ijisa.2012.09.03Copyright©2012MECSI.J.IntelligentSystemsandApplications,2012,9,22-28COCOMOEstimatesUsingNeuralNetworksAnupamaKaushik,AssistantProfessor,Dept.ofIT,MaharajaSurajmalInstituteofTechnology,GGSIPUniversity,Delhi,Indiaanupama@msit.inAshishChauhan,DeepakMittal,SachinGuptaDept.ofIT,MaharajaSurajmalInstituteofTechnology,GGSIPUniversity,Delhi,IndiaAshish.chauhan004@gmail.com;deepakm905@gmail.com;sachin.gupta_15@yahoo.comAbstract—Softwarecostestimationisanimportantphaseinsoftwaredevelopment.Itpredictstheamountofeffortanddevelopmenttimerequiredtobuildasoftwaresystem.Itisoneofthemostcriticaltasksandanaccurateestimateprovidesastrongbasetothedevelopmentprocedure.Inthispaper,themostwidelyusedsoftwarecostestimationmodel,theConstructiveCostModel(COCOMO)isdiscussed.Themodelisimplementedwiththehelpofartificialneuralnetworksandtrainedusingtheperceptronlearningalgorithm.TheCOCOMOdatasetisusedtotrainandtotestthenetwork.ThetestresultsfromthetrainedneuralnetworkarecomparedwiththatoftheCOCOMOmodel.TheaimofourresearchistoenhancetheestimationaccuracyoftheCOCOMOmodelbyintroducingtheartificialneuralnetworkstoit.IndexTerms—ArtificialNeuralNetwork,ConstructiveCostModel,PerceptronNetwork,SoftwareCostEstimationI.IntroductionSoftwarecostestimationisoneofthemostsignificantactivitiesinsoftwareprojectmanagement.Accuratecostestimationisimportantbecauseitcanhelptoclassifyandprioritizedevelopmentprojectstodeterminewhatresourcestocommittotheprojectandhowwelltheseresourceswillbeused.Theaccuracyofthemanagementdecisionswilldependontheaccuracyofthesoftwaredevelopmentparameters.Theseparametersincludeeffortestimation,developmenttimeestimation,costestimation,teamsizeestimation,riskanalysis,etc.Theseestimatesarecalculatedintheearlydevelopmentphasesoftheproject.So,weneedagoodmodeltocalculatetheseparameters.Anearlyandaccurateestimationmodelreducesthepossibilitiesofconflictsbetweenmembersinthelaterstagesofprojectdevelopment.Inthelastfewdecadesmanysoftwarecostestimationmodelshavebeendeveloped.Thealgorithmicmodelsalsoknownasconventionalmodelsuseamathematicalformulatopredictprojectcostbasedontheestimatesofprojectsize,thenumberofsoftwareengineers,andotherprocessandproductfactors[1].Thesemodelscanbebuiltbyanalysingthecostsandattributesofcompletedprojectsandfindingtheclosestfitformulatoactualexperience.COCOMO(ConstructiveCostModel),isthebestknownalgorithmiccostmodelpublishedbyBarryBoehmin1981[2].Itwasdevelopedfromtheanalysisofsixtythreesoftwareprojects.Theseconventionalapproacheslacksintermsofeffectivenessandrobustnessintheirresults.Thesemodelsrequireinputswhicharedifficulttoobtainduringtheearlystagesofasoftwaredevelopmentproject.Theyhavedifficultyinmodellingtheinherentcomplexrelationshipsbetweenthecontributingfactorsandareunabletohandlecategoricaldataaswellaslackofreasoningcapabilities[3].Thelimitationsofalgorithmicmodelsledtotheexplorationofthenon-algorithmicmodelswhicharesoftcomputingbased.Nonalgorithmicmodelsforcostestimationencompassmethodologiesonfuzzylogic(FL),artificialneuralnetworks(ANN)andevolutionarycomputation(EC).Thesemethodologieshandlereallifesituationsbyprovidingflexibleinformationprocessingcapabilities.ThispaperproposedaneuralnetworktechniqueusingperceptronlearningalgorithmforsoftwarecostestimationwhichisbasedonCOCOMOmodel.Neuralnetworkshavebeenfoundasoneofthebesttechniquesforsoftwarecostestimation.Now-a-daysmanyresearchersandscientistsareconstantlyworkingondevelopingnewsoftwarecostestimationtechniquesusingneuralnetworks[4,5,6,7].Therestofthepaperisorganizedasfollows:section2and3describestheCOCOMOmodelandartificialneuralnetworkconceptsrespectively.Section3and4discussestherelatedworkandproposedneuralmodel.Section4and5presentstheproposedmodelandthetrainingalgorithmimplemented.Section6discussestheexperimentalresultsandevaluationcriteria.FinallySection7concludesthepaper.II.COCOMOModelTheCOCOMOmodel[2]isthemostwidelyusedalgorithmiccostestimationtechniqueduetoitsCOCOMOEstimatesUsingNeuralNetworks23Copyright©2012MECSI.J.IntelligentSystemsandApplications,2012,9,22-28simplicity.Thismodelprovidesuswiththeeffortinpersonmonths,thedevelopmenttimeinmonthsandtheteamsizeinpersons.Itmakesuseofmathematicalequationstocalculatetheseparameters.TheCOCOMOmodelisahierarchyofsoftwarecostestimationmodelsandtheyare:1.BasicModel-ItestimateseffortforthesmalltomediumsizedsoftwareprojectsinaquickandroughfashionandtakestheformE=a(SIZE)b(1)whereEiseffortappliedinPerson-MonthsandSIZEismeasuredinthousanddeliveredsourceinstructions.Thecoefficientsaandbaredependentuponthethreemodesofdevelopmentofprojects.Boehmproposedthreemodesofprojects:(a)Organicmode-Itisforsmallsizedprojectsupto2-50KLOC(thousandlinesofcode)withexperienceddevelopersinafamiliarenvironment.(b)Semidetachedmode-Itisformediumsizedprojectsupto50-300KLOCwithaveragepreviousexperienceonsimilarprojects.(c)Embeddedmode-Itisforlargeandcomplexprojectstypicallyover300KLOCwithdevelopershavingveryl