Customer Chrun Prediction Using Ensemble Learning In Telecommunication Industry

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-005

Volume: 13 Issue: 04 | Apr 2026 www.irjet.net p-ISSN: 2395-0072

Customer Chrun Prediction Using Ensemble Learning In Telecommunication Industry

Ms. Riteeka Rajubhai Patel1, Dr. Rachna Mukesh Patel2

1 PG Student, Department of Computer Engineering (M. Tech Computer Engineering)

2 Assistant Professor, Department of Computer Engineering, Chhotubhai Gopalbhai Patel Institute of Technology, Bardoli, 394 350, India

Abstract - The importance of Customer Churn Prediction (CCP) hasincreasedlatelyparticularlyintelecommunication industries. Different versions of Churn Prediction Models (CPMs) have been designed by different authors. Technologies like machine learning, data mining and other meta-heuristic algorithms have been used in these models. The significant churn prediction methods designed over the recent years have been discussed in this paper. The purpose of this study is to evaluate the customer survival and customerthreatscenarioswhichresultinunderstandingthe causeof churnsintelecommunicationbusinesses.Theclients whoareabouttoturnintochurnsarealsoexaminedandthe time they take to turn into churns is also calculated. The causes of customer churn and the behavior of churn are highlighted through this study. To have a better understanding of the churn prediction, this research summarizesthe variouschurnpredictionmethods.Inmostof the models, instead of using individual algorithms, hybrid models are designed. Thus, the telecommunication businesses can improve their services towards the high-risk clients so that the decision of clients to opt for churn is avoided. The voting classifier and SVM classifier are implemented for the churn prediction. The performance of voting classifier and SVMclassifier is compared in terms of accuracy, precision and recall. It is analyzed that accuracy, precision and recall of voting classifier is high as compared to SVM classifier.

Key Words: Churn Prediction, Data Mining, Telecommunication Industry

1. INTRODUCTION

Machine learning is a branch of artificial intelligence that allows computer systems to learn directly from examples, data, and experience. Through enabling computers to perform specific tasks intelligently, machine learning systemscancarryoutcomplexprocessesbylearningfrom data, rather thanfollowing pre-programmedrules. Recent yearshaveseenexcitingadvancesinmachinelearning,which have raised its capabilities across a suite of applications. Increasing data availability has allowed machine learning systems to be trained on a large pool of examples, while increasing computer processing power has supported the analyticalcapabilitiesofthesesystems.Withinthefielditself therehavealsobeenalgorithmicadvances,whichhavegiven machine learning greater power. As a result of these

advances,systemswhichonlyafewyearsagoperformedat noticeablybelow-humanlevelscannowoutperformhumans at some specific tasks. Many people now interact with systemsbasedonmachinelearningeveryday,forexamplein image recognition systems, such as those used on social media;voicerecognitionsystems,usedbyvirtualpersonal assistants;andrecommendersystems,suchasthoseusedby online retailers. As the field develops further, machine learning shows promise of supporting potentially transformativeadvancesinarangeofareas,andthesocial and economic opportunities which follow are significant. Machinelearningalgorithmsaregenerallycategorizedinto unsupervised,supervised,andsemi-supervisedlearning.If instances are given with known labels (the corresponding correct outputs) then the learning is called supervised, in contrast to unsupervised learning, where instances are unlabelled.Semi-supervisedlearningcombinesbothlabeled andunlabeledexamplestogenerateanappropriatefunction orclassifier.

Today’s customers have infinite numbers of data sources. Smartphoneallowsaquickeraccesstotheproduct, brandandprice-comparisondata.Consequently,companies inseveralindustriesaretakingstrivingtoattractrecollecting customers.Becauseofquicktechnologicalimprovementsand improvedcompetition,customershaveseveraloptions.The tasks have been developed from telecommunication operators.Companiesaredroppingmanyrevenuesbecause the clients switch their interests and choose other service providers.Thisprocedureisknownas“Churn”.

Churn is one of the most significant administration features in telecommunication industry [1]. Churn can be definedasabroadertermthatmightincludeseveralactions like the customer’s service is done through customer themselvesorbyaserviceproviderthroughdiscreteservice agreements [2], [3]. But, the most important and the most commonreasonofchurnisthenon-satisfactionofcustomer towardsanyserviceaccessedthroughaprovider[4],[5].Yet thesedonotsimplycause.Itisdescribedthatseveralissues have been arising due to the customer churn because of increase in competition, saturated markets, dynamic conditions, and beginning novel attractive customers. Frequently, client started churn is complicated and issues relatedtoeachchurnmightvaryforeverycustomer.Thus, this study emphasizes on studying the various types of

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-005

Volume: 13 Issue: 04 | Apr 2026 www.irjet.net p-ISSN: 2395-0072

churnsandthedifferentmethodsproposedasasolutionto avoidtheirissues.

In telecommunication businesses, both, voice and informationservicecustomerschooseservice-providerbya huge range of businesses and have self-determination towardsswitchingtheirprivilegesfromoneserviceprovider to another which can lead to improvement. There is huge growthinsuchcompetition.Theclientsdemandtoplevelof products and services at significantly reduced prices [6]. Several telecommunication companies use retention approaches [7] towards coordinate services to save customers through extensive tenure. In such case, a key businesshasbeendevelopedwiththeaimofreducingchurn. Reducing the churn is proposed for the provisional telecommunication corporations. To avoid huge losses for clients and to evaluate the time of churn in sequence of including their requirements, this step is important [8]. In common,severaldataminingmethods[9],[10]areemployed to forecast churns like there are approaches which are associated with sentiment analysis. Machine learning and meta-heuristicmethods[11]arealsoimportantlypresented aschurnpredictionapproaches.

Churn is an issue caused by the customers who are furtherexclusiveinobtainingnewclients.Clientrevocation should be specified while client has been clogged by their SIMcard.Whenthewordchurnismentioned,theonething whichcomesinmindisthatcustomerchurnwhichisamajor issueofthetelecommunicationmarket[12].Churnintends those customers who want to leave in the nearby future. Thereisessentialneedtopredictthosecustomersonbehalf of some parameter to initiate some suitable actions for minimizing their leaving. Most of the mobile phone companies invest under CRM (Customer Relationship Management)technology[13][14].

2. Churn Prediction Techniques Methodology

The greatest churn prediction techniques are the metaheuristicalgorithmswhicharedesignedfromgreatlyprecise prediction. Following are the techniques, algorithms, and methodologiesthatareusedbydifferentresearchersintheir research.

2.1 Decision Tree (DT)

Decision Tree is maximum prominent predictive model thatisusedforthepurposeofclassificationofupcomingtrial [20], [21].It comprises two stages, tree pruning and tree building. In tree building training set data is recursively partitionedinaccordancewiththevaluesoftheattributes. Thisprocedureservesonuptothereisnoonepartitionis lefttohaveidenticalvalues.Inthisprocesssomevaluesmay beremovedfromthedataduetonoisydata.Majorevaluated error rate branches are selected and then unconcerned in

pruning.Towardspredictaccuracyandreducingcomplexity ofthedecisiontreeiscalledtreepruning[16],[17].

2.2 Linear Regression Model (LRM)

To predict customer satisfaction, the regression analysis model is another popular technique that is based on supervisedlearningmodel.Inthismodeladatasetofpast observationsisusedtoseefuturevaluesofexplanatoryand numerical targeted variables [17]. The formula of LRM is given[18].

yisabinaryvariable.Thisshowsanevent.Ify=1theevent occurselsenotoccur.

X1,X2,……,XKbetheself-determininginputs.

B0,B1,….,BK bethefailure.

2.3 Naive Bayes Model (NBM)

Inthismodel,theprobabilitiesofspecifiedinputsampleare calculatedthatgoestowardsaparticularclassy.Thesetof variable is given (X1 Xn).The given formula is used to calculateprobabilities[19].

The yj is the probability of the previous calculations. The probabilityofindependentvariableisindependent.

2.4 Neural Networks Model (NNM):

TheNeuralNetworksModelisusedtoelaboratefunctionality likenon-linear.Themodelholdsthecapabilitytolearndueto itscomparabledataprocessingstructure.Thesetechniques providesuccessfulresultsafterapplyingtomanyproblems like classification, control, and prediction due to the biologicalbrain[15].Themodelisdissimilartoclassification model as well as decision tree due to its likely hood prediction. The neural network has several techniques havingmeritsanddemerits.Theresearchersuggestsneural network is well than decisiontreeandregressionanalysis modelofchurnprediction[16].

2.5 Support Vector Machine (SVM)

The Support Vector Machine classifier deals with a linear permutation of subset of the training set by finding a maximumedgeover-energizedplane.TheSVMplotsthedata intohighdimensionalfeaturesspaceclosingtoinfinitewith the help of most important part if vectors are nonlinearly

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-005

Volume: 13 Issue: 04 | Apr 2026 www.irjet.net p-ISSN: 2395-0072

divisibleinputfeatures[22]andthencategorizeinformation throughmaximumscopehyper-plane.

Where,

M=No.ofsamplesintoatrainingdataset.

Xi=Vectorsupportwhenai>0

/=Corefunction

X=Unidentifiedsamplefeaturevector

d=Doorstep.

(ai)is a parameter that is the result of curved quadratic programmingproblemswithrespecttolinearconstraint[23].

2.6 Fuzzy Logic Algorithm:

Afuzzylogictechniqueisverysimpletounderstanddueto its very simple mathematical concepts and fuzzy reasons. Fuzzy logic has the property of flexibility, tolerant of indefinite data. The function of random data can be implementedinthis model. Inmost of the cases,thefuzzy logic system spends the idea of the predictable managed techniques and streamlines the operations. In telecommunication industries no work has been achieved relatedtochurnpredictionwithfuzzylogictechniques[24].

2.7 Evolutionary Learning Data Mining Techniques

Evolutionarylearningdataminingtechniquesarestochastic searchalgorithmswhichareinspiredbytheprocessofneoDarwinianevolution.Dataminingbyevolutionarylearning technique is inherited classification techniques. The motivationforapplyingevolutionarylearningtodatamining is that they are robust, adaptive search techniques that performaglobalsearchinthesolutionspace[45].Suchtypes of genetic algorithms have some set of rules. It creates a series of random rules to be checked against a training dataset.Theruleswhichmostcloselyfitthedataareselected andaremutated.Thesetechniquesapplytheserulesonsome givendatasetthatprovidedecision-makingresults[25].

2.8 K-means clustering

Themostwell-knownandrelevanttechniqueofclusteringis K-MeanspresentedbyMc.Queenin1967.Thefollowingare themainstepsinKmeansclustering.InKmeanclusterin approach,inthefirststepweselectkobjectsthathavetheir center(mean).Inthismethodtheremainingobjectsarenot selected yet are assigned to cluster with respect to the similarity of the object with cluster. These similarities are measuredonthebehalfofthedistancebetweenclustermean andobjectandafterthiscalculation,thenewcenterpointis

calculatedon the behalfofabove factand werepeatthese steps until the required function is achieved. In k mean clustering,themostimportantpointistofindthenumbersof clustersthatare optimumaswell asthedistance between clustermeanandobjects.Thealgorithmworksuntilnonew clusterelementleaveaclusterandenterintoothercluster and no new center point is set for any cluster. When this targetisachievedthealgorithmisstopped[26].

2.9 Ant Colony Optimization (ACO)

Ant colony optimization met empirical motivated seeking performance of actual ant colonies [27].An algorithm is a practical behavior of actually living ant that is an insect havingsomerulesusedbythemtofindthefoodfromhisnest through shortest path first towards food source. 1st Ant colonyalgorithmwasdesignedasantsystem[28].Antcolony optimization has workforce artificial ants works like biologicalanttofindtheoptimumsolution.InAntsystemin first step an ant selects a path to reach a point we set pheromonevaluebutincaseofproblem,aheuristicvalueis set.Thepheromonevalueshowsthetrailandheuristicvalue showstheproblems.Antcolonyoptimizationisappliedtoa largecollectionofproblems[27],[30],likevehiclerouting problem, scheduling [29] and routing in packet-switched networks[31,32]inrecenttimes.Antcolonyoptimization hasappliedunderdataminingfield[33].

3. LITERATURE REVIEW

J.Buresetal,[33]investigatedthewaysthroughwhich the class imbalance of churn prediction could be handled more appropriately. Inspecting (random and progressive under-sampling),cost-sensitivelearner (weighted random forests (WRF)) and boosting (gradient boosting machine (GBM)) are utilized by adjusting for calculating churn expectation precision. An improvement in the prediction accuracywasachievedbyapplyingunder-sampling.

Veronika Effendy et al, [34] anticipated proficient systemthrough the imbalancedinformationtaking care of issueofupgradingclientchurnexpectation.Plannedmethod integratestheexaminationbyadjustingthedatasetwiththe aim of upgrading churns forecast exactness. Examining procedureisitselfblendingofunder-inspectingandSMOTE (Synthetic Minority Oversampling Technique). The core procedureincludesexaminationbyirregularityinformation issuewhenWRFgroupsdatausingspecificchurnprediction. Combined inspecting procedure expands F-measure & precision esteems demonstrating decrease of information records through exact forecast. Despite the fact that exhibitionisentirelygreatitisnotmuchbeneficialtoutilize commonunder-samplingplan.

NingLuetal,[35]proposedamethodinwhichboosting was applied to improve a client churn prediction model. Dependingupontheweightassignedbyboostingalgorithm, theclientswereseparatedintotwoclusters.Thisresultsin identifyingahigherriskclientcluster.Asabasicleaner,the

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-005

Volume: 13 Issue: 04 | Apr 2026 www.irjet.net p-ISSN: 2395-0072

logistic regression was applied in this research. On each clusterachurnpredictionmodelwasdesigned.Experiments wereconductedandacomparativeanalysisofproposedand a single logistic regression model was presented. The outcomes showed that for churn prediction analysis, boostingprovidedhighlyefficientresults.XiaojunWuetal, [36]suggestedapredictionstrategydependentonenhanced SMOTEandAdaBoostbyanticipatinginternetbusinessclient churn. In this methodology, at first better SMOTE is connectedinordertoprocessunevendatasetsusingblendof over-examining & under-sampling. At that point decent datasetispreparedintoAdaBoostlearningalgorithmusing weakclassifiertowardsorderclients&anticipatechurn.G. GaneshSundarkumaretal,[37]plannedone-classSVMbuilt under-inspecting by improving churn & protection fraud discovery. At first, information is under sampled utilizing one-classSVM andafterthat;groupingisperformedusing machine learning algorithms. In view of outcomes it is inferred that decision tree performs superior towards furtherclassificationalgorithmsandalongsideone-classSVM it reduces system complexity nature & increases forecast precision. Qiu Yihui et al, [38] recommended that current CCPtechniquesdon'thavemanylogical,systemtheoryand strategy so it is unable to fulfill application requirements. Creators planned an element determination technique dependent on oriented ordering pruning method. This methodology provided the pruning question of classifier blendratherthanattributechoice.Insubsequentadvance,an element extraction technique is planned for extricating various aspects by higher request client information. By assessment results it is discovered that proposed method improves the churn expectation by oriented ordering pruningmethod.QiuhuaShenetal,[39],usedfeature-based churnexpectationdevelopmentandplannedframeworkof corresponding combination of multilayer includes for increasingspreadchurnforecastrate.Suggestedframework utilized element factorization and aspect developmentby combination of features. This methodology builds churn predictionexactnessbysettlinghighdimensionsanduneven informationissue.Bethatasitmay,elementdetermination procedureisadequateandconsequentlyuneveninformation issuereappears.

SebastiánMaldonadoetal,[40]suggestedaproductive featuredeterminationtechniqueutilizingSVMdependenton thebenefitsmodel.Itsmethodologycentersonchoosingbest using the classifier stage. SVM classifier is built on benefit premisewhencomponentfactorsarealsochosenbythought of benefit. Methodology adaptable permits bit capacities using the enhanced prediction accuracy. Administrative causes are not fulfilled into SVM like base classifier.Aimee Backiel et al, [41] recommended utilization of a blend of nearby and social features using churn prediction as 2 element models distinguish distinctive arrangement of churners. An outfit approach is utilized by joining two features. Client information and social information of cell phone specialist organizations are utilized for assessment. Proposedmodelcomprisesspreadingenactmentcalculation

which spreads nearby and social factors between social & neighborhoodmodel&gatheringmodeltoconsolidatethese aspects together. Result of assessment infers that churn prediction is better while utilizing consolidated model of aspectsasopposedtoutilizingindividualmodelsorelement models. The principle confinement of this methodology is that exclusion of non-client nodes in production of call diagrambecauseofbiggervolumeofinformationdecreases adequacyofchurnprediction.Anadditionaldisadvantagein this methodology is exclusion of bad energies by social network.Social Network analysis may improve CCP as suggestedbyAiméeBackieletal,[42].Authorsproposedthe frameworkthatintegratessocialnetworkinformationinto collective churn prediction model by local and real-time attributes.Estimationoutcomeshowsthatchurnprediction is enhanced in terms of accuracy, AUC (Area Under ROC Curve)andliftpercentage.

Pretam Jayaswal et al, [43] recommended ensemble method through forecast of churn. Recommended methodologyutilizedclientuseandrelateddataonethrough investigation of telecommunication client churn. Decision tree and its troupes, Random forest and Gradient boosted trees are used for structure of binary churn classifier. Assessment demonstrates that these ensemble-based methodologiesparticularlytheinputimprovementbuiltGBT grouphas betterexactness andaffectabilitythroughclient beatexpectation.Thus,methodologyisn'ttriedonongoing information and this confines unwavering quality on this model.AnujSharmaetal.[17],Marketingwritingstatesthat it is more exorbitant to draw in another customer than to holdacurrentsteadfastcustomer.Churnpredictionmodels are created by scholastics and specialists to adequately overseeandcontrolcustomerchurnkeepinginmindtheend goaltoholdexistingcustomers.Aschurnmanagementisan imperativeactionforcompaniestoholdfaithfulcustomers, the capacity to accurately anticipate customer churn is fundamental.Asthecellnetworkadministrationsadvertise windingupmorefocused,customerchurnmanagementhas turned into a vital errand for mobile communication administrators.Thispaperproposesaneuralnetwork(NN) based way to deal with foresee customer churn in membershipofcellremoteadministrations.AdemKarahoca etal.[8],churnmanagementisessentialandbasicissuefor Global Services of Mobile Communications (GSM) administratorstocreateproceduresandstrategiestokeep itsendorsersofpassotherGSMadministrators.Inthefirst place period of churn management begins with profile creation for the endorsers. Profiling process assesses call detail information, money related data, calls to customer benefit,contractpointsofinterest,showcasesubtleelements andgeographicandpopulaceinformationofagivenstate.In this examination, input features are clustered by x-means and fuzzy c-means clustering algorithms to put the supporters into various discrete classes. Adaptive Neuro Fuzzy Inference System (ANFIS) is executed to build up a delicatepredictiondemonstrateforchurnmanagementby utilizingtheseclasses.

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-005

Volume: 13 Issue: 04 | Apr 2026 www.irjet.net p-ISSN: 2395-0072

Author Dataset

AdemKarahoca[8]

ClementKirui[9]

Features

GSMoperator, Turkey24,900 customers22 attributes

Demography, Usagepattern, Valueadded services

Europeanoperator 106,405customers 112attributes

Ballings,Michel[10]

Ismail, Mohammad [11]

Unknown129,892 customers113 attributes

Unknown,169 customers10 attributes

Contract,usage pattern patterns,and callspattern

Demographic, Valueadded, usagepattern

x-Means clustering, Adaptive NeuroFuzzy Inference System

NaïveBayes, Decision Tree

Precision andRecall

Confusion matrix, accuracy, precision, recall

Logistic regression, Bagging, Decision Tree AUC

Demographic, Billingdata, usagepattern, customer relationship Neural network, Regression

Confusion matrix, accuracy, precision, recall

Theprecision andrecall valuesare achieved Thecomplexity ofthesystemis high

Thedecision treemethod willarrange data efficiently

Thebagging classification methodgive goodaccuracy

Thedatasetis collectedin realtime

Thenaïvebayes reduce effectivenessof thesystem

Precision,recall canbe calculated

Regression natureincrease executiontime

HLee[12]

Cell2CellDataset 100,000customers 171attributes

Behavioral information, Customercare and demographics

Stepwise variable selection partialleast squares

Proportion ofhit records

Thebehavior ofthe customersare analyzed

The classification canbeapplied infuture

AnujSharma[17]

MLDatasetatUCI 2,427customers20 attributes

Demographics, Usagepattern, Valueadded services

Artificial Neural Network

Confusion matrix

Theresults are performedon highrange dataset

Thetraining timeofthe systemishigh

AbbasKeramati[18]

Iranian telecommunication operator3150 customers15 attributes

Demographic, callusage pattern, customercare service

Binomial logistic regression model

Statistical hypothesis test

Thepatterns ofthecallsare calculated

Statistical methodis appliedforthe analysis

KristofCoussement [19]

Belgian134,120 customers27 attributes

Demographic Usagepatter, billand payment

Generalized additive models (GAM)

AUCtopdecilelift

Featuresare extracted efficiently Theaccuracyis low

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-005

Volume: 13 Issue: 04 | Apr 2026 www.irjet.net p-ISSN: 2395-0072

Author Dataset

MarcinOwczarczuk [20]

Umayaparvathi[21]

Polishmobile operator122098 customers1381 attributes

Cell2CellDataset 100,000customers 171attributes

Demographic, calldata records, customercare services Logistic regression Decisiontree Liftcurves

Thedatasetis collectedin realtime Theaccuracy canbe increased

Behavioral information, Customercare and demographics

Gradient Boosting, Decision Tree, Support Vector Machine, Random Forest,KNN,Ridge Regression& Logistic Regression

Confusion matrix, accuracy, precision, recall,F1score

Thenumber ofclassifiers areapplied forthe prediction analysis

Thecomplexity ofthesystemis quitehigh

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-005

Volume: 13 Issue: 04 | Apr 2026 www.irjet.net p-ISSN: 2395-0072

3. RESEARCH METHODOLOGY

Thisresearchworkisbasedonthechurnpredicationusing techniques of machine learning. The churn prediction techniques have various phases which include preprocessing, feature extraction and classification. The datasetofthechurnpredictioniscollectedfromthekaggle. The kaggle is the authentic source for the dataset collection. The collected data has various missing and redundantvalueswhichareremovedtocleanthedataset. In the second phase, features are extracted from the dataset. The feature extraction approach will establish relationshipbetweeneachattributeandtargetset.Inthe last phase the hybrid method is applied for the churn prediction.Thehybridmethodwillbethecombinationof the KNN classifier and decision tree classifier. The KNN classifier will extract the features and decision tree will generatefinalresults.Thevotingclassifieristheproposed algorithmwhichisusedforthechurnprediction.

START

Inputdatasetforthechurn prediction

Applypre-processingphaseto removemissingandredundant values

ApplyPCAAlgorithmforfeature reduction

ApplyKNN algorithm

ApplyDecision Treealgorithm

Preparetrainingsetbased onbothclassifiers

Applyvotingclassifierfor theprediction InputTest Set

Analyzeperformancein termsofaccuracy,precision, recall

STOP

4. RESULT AND DISCUSSION

Thedatasetiscollectediscollectedfromthekaggleandthe results are analyzed in terms of accuracy, precision and recall.

Importantmetricsconsideredtoanalysetheefficiencyof thesealgorithmsinclude:

1. Precision: Precision is the degree to which repeated measurements under static conditions generate similar outcomes.

2. Recall: It is ratio of properly predicted positive observationstotheallobservationsinoriginalclass.

3. Accuracy: It is the ratio of the accurately labelled subjectstotheentiregroupofsubjects.

Table1:PerformanceAnalysis

Figure4.5:PerformanceAnalysis

Asdepictedinfigure4.5,theefficiencyofexistingandnew

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-005

Volume: 13 Issue: 04 | Apr 2026 www.irjet.net p-ISSN: 2395-0072

algorithmiscomparedwithrespecttocertainmetrics.Itis analysedthatpercentageofallthreeparametersishigher inthenewalgorithmicapproach.

3. CONCLUSIONS

Customerchurnhasbeenacceptedlikeimportantconcern in to aggressive research and the telecommunication industryhasbeenengagedinthisbyrelatingseveraldata mining strategies. Various data mining approaches are generally useful in customer churn. Telecommunication trade has proficient great churn rates and gigantic churningmisfortune.Inunkindnessofpointthatindustry troubleisinescapable,howeveratsimilartimechurnmay beadministeredandsavedatacceptablelevel.Thispaper appraised diverse classifications of client information available into open datasets, predictive models and performancemetricsusedsuchportionofwritingthrough churn prediction into telecommunication industry. The churn prediction has various phases which include preprocessing,featureextractionandclassification.TheSVM and voting classifiers are implemented for the churn prediction. It is analyzed that voting classifier have high accuracy,precisionandrecallforthechurnpredictionas comparedtoSVMclassifier.

REFERENCES

[1] Liao, Shu-Hsien, Pei-Hui Chu, and Pei-Yuan Hsiao. "Data mining techniques and applications–A decade review from 2000 to 2011." Expert Systems with Applications39,no.12(2012):11303-11311.

[2] Kamalraj, N., and A. Malathi. "A survey on churn predictiontechniquesinthecommunicationsector." InternationalJournalofComputerApplications64,no. 5(2013).

[3] N.Hashmi, N.ButtandM.Iqbal. Customer Churn Prediction in Telecommunication A Decade Review andClassification.InternationalJournalofComputer ScienceVol.10(5),2013

[4] V. Umayaparvathi, K. Iyakutti, “ Applications of Data Mining Techniques in Telecom Churn Prediction”, International Journal of Computer Applications, Vol. 42,No.20,2012

[5] V.Umayaparvathi,K.Iyakutti,“AttributeSelectionand Customer Churn Prediction in Telecom Industry”, ProceedingsoftheIEEEInternationalConferenceOn DataMiningandAdvancedComputing,2016(tohave appeared).

[6] Huang,Bingquan,MohandTaharKechadi,andBrian Buckley. "Customer churn prediction in telecommunications." Expert Systems with Applications39,no.1(2012):1414-1425

[7] Shaaban, Essam, Yehia Helmy, Ayman Khedr, and Mona Nasr. "A proposed churn prediction model." IJERA2(2012):693-697

[8] Adem Karahoca, Dilek Karahoca,”GSM churn management by using fuzzy c-means clustering and adaptive neuro-fuzzy inference system”, Expert SystemswithApplications38(2011)1814–1822.

[9] Kirui,Clement,LiHong,WilsonCheruiyot,andHillary Kirui."Predictingcustomerchurninmobiletelephony industryusingprobabilisticclassifiersindatamining." IJCSIInternationalJournalofComputerScienceIssues 10,no.2(2013):1694-0784.

[10]Ballings, Michel, and Dirk Van den Poel. "Customer eventhistoryforchurn prediction:Howlongislong enough?."ExpertSystemswithApplications39,no.18 (2012):13517-13522

[11]Ismail, Mohammad Ridwan, Mohd Khalid Awang, M. Nordin A. Rahman, and Mokhairi Makhtar. "A MultiLayerPerceptronApproachforCustomerChurn Prediction."International Journal of Multimediaand UbiquitousEngineering10,no.7(2015):213-222.

[12]HLee, YLee,HCho, KIm,YSKim,"Miningchurning behaviorsanddevelopingretentionstrategiesbased on a partial least squares (PLS) model", Decision Support Systems, Volume 52, Issue 1, 2011, Pages 207–216

[13]Ondrus J, Pigneur Y. Coupling mobile payments and CRMintheretailindustry;2004.p.1–8.

[14]Gotovac S. Modeling data mining applications for prediction of prepaid churn in telecommunication services.2010;51(3):275–83.

[15]Behara R, Fisher W, Lemmink J. Modelling and evaluatingservicequalitymeasurementusingneural networks. International Journal of Operations and Production Management. 2002; 22(10):1162–185. Crossref

[16]Chan C, Yao X. A novel evolutionary data mining algorithmwithapplicationstochurnprediction.2003; 7(6):532–45.

[17]Anuj Sharma, Dr.Prabin Kumar Panigrahi, "A Neural Network based Approach for Predicting Customer Churn in Cellular Network Services", International JournalofComputerApplications,Volume27–No.11, 2011,pp.0975–8887.

[18]AbbasKeramati,SeyedM.S.Ardabili,"Churnanalysis foranIranianmobileoperator",Telecommunications Policy,35,2011,pp.344–356.

[19]KristofCoussement,DriesF.Benoit,DirkVandenPoel, "Improvedmarketingdecisionmakinginacustomer churn prediction context using generalized additive models", Expert Systems with Applications, Volume 37,Issue3,2010,Pages2132–214.

[20]Marcin Owczarczuk, "Churn models for prepaid customersinthecellulartelecommunicationindustry using large data marts", Expert Systems with Applications,37,2010,pp.4710–4712.

[21]V.Umayaparvathi,K.Iyakutti,,“AttributeSelectionand Customer Churn Prediction in Telecom Industry”, ProceedingsoftheIEEEInternationalConferenceOn Data Mining and Advanced Computing, 2016 (to be appeared).

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-005

Volume: 13 Issue: 04 | Apr 2026 www.irjet.net p-ISSN: 2395-0072

[22]BurgesCJC.Atutorialonsupportvectormachinesfor patternrecognition.1998;2:121–67.

[23]Data Mining and Knowledge Discovery. Springer. 1997;2(2):121–67.

[24]SelviO.Trafficaccidentpredictionsbasedonafuzzy logic approach for safer urban environments. Case Study:IzmirMetropolitanArea;2009.p.1–131.

[25]Dorigo M, Stützle T. Ant colony optimization. Cambridge.MA:MITPress;2004.p.1–321.

[26]Abhishek Kumar K1 and Sadhana2SURVEY ON KMEANSCLUSTERINGALGORITHM

[27]Dorigo M, Maniezzo V, Colorni A. Ant system optimizationbyacolonyofcooperatingagents.IEEE TransactionsonSystems,Man,andCybernetics-PartB Cybernetics. 1996; 26(1):29–41. Crossref. PMid:18263004

[28]Larivière B, Van D, Poel, D. Predicting customer retentionandprofitabilitybyusingtherandomforest andregressionforesttechniques.ExpertSystemswith Applications.2005;29(2):472–84.Crossref

[29]Vanthienen J, Mues C, Wets G, Delaere K. A toolsupported approach to inter-tabular verification. ExpertSystemswithApplication.1998;15(3):277–85. Crossref

[30]BabuS,AnanthanarayananNR,RameshV.Asurveyon factors impacting churn in telecommunication using data mining techniques. International Journal of Engineering Research and Technology. 2014; 3(3):1745–8.

[31]Abraham A, Ramos V. Web usage mining using artificial ant colony clustering. Congress on EvolutionaryComputation;2003.p.1384–91.Crossref

[32]HandlJ,KnowlesJ,DorigoM.Ant-basedclusteringand topographicmapping.Artificial Life.2006;12(1)35–61.Crossref.PMid:16393450

[33]Churi A, Mayuri D, Sonal D, Prajakta K. Analysis of customer churn in the mobile industry using data mining.InternationalJournalofEmergingTechnology andAdvancedEngineering.2015;5(3):2250–459.

[34]]. J. Burez and Dirk Van den Poel. "Handling class imbalance in customer churn prediction." Expert Systems with Applications, vol. 36, no. 3, pp. 46264636,2009.

[35]Veronikha Effendy and ZK AbdurahmanBaizal. "Handling imbalanced data in customer churn prediction using combined sampling and weighted randomforest."In20142ndInternationalConference on Information and Communication Technology (ICoICT),pp.325-330.IEEE,2014.

[36]Ning Lu, Hua Lin, Jie Lu, and Guangquan Zhang. "A customer churn prediction model in the telecom industry using boosting." IEEE Transactions on Industrial Informatics, vol. 10, no. 2, pp. 1659-1665, 2014.

[38]G. Ganesh Sundarkumar, Vadlamani Ravi, and V. Siddeshwar."One-classsupportvectormachine-based undersampling: Application to churn prediction and insurancefrauddetection."In2015IEEEInternational Conference on Computational Intelligence and ComputingResearch(ICCIC),pp.1-7.IEEE,2015.

[39]Qiu Yihui, and Zhang Chiyu. "Research of indicator systemincustomerchurnpredictionforthetelecom industry." In201611thInternational Conferenceon ComputerScience&Education(ICCSE),pp.123-130. IEEE,2016.

[40]Qiuhua Shen, Hong Li, Qin Liao, Wei Zhang, and KoneKalilou. "Improving churn prediction in telecommunicationsusingacomplementaryfusionof multilayer features based on factorization and construction."The26thChineseControlandDecision Conference(2014CCDC),pp.2250-2255.IEEE,2014.

[41]Sebastián Maldonado, Álvaro Flores, Thomas Verbraken, Bart Baesens, and Richard Weber. "Profitbased feature selection using support vector machines–Generalframeworkandanapplicationfor customerretention."AppliedSoftComputing,vol.35, pp.740-748,2015.

[42]AiméeBackiel,YannickVerbinnen,BartBaesens,and GerdaClaeskens."Combininglocalandsocialnetwork classifiers to improve churn prediction." In 2015 IEEE/ACM International Conference on Advances in SocialNetworksAnalysisandMining(ASONAM),pp. 651-658.IEEE,2015.[

[43]Aimée Backiel, Bart Baesens, and GerdaClaeskens. "Predictingtime-to-churnofprepaidmobiletelephone customers using social network analysis." Journal of theOperationalResearchSociety,pp.1-11,2016.

[44]Pretam Jayaswal, BakshiRohit Prasad, DivyaTomar, and SonaliAgarwal. "An Ensemble Approach for Efficient Churn Prediction in Telecom Industry." International Journal of Database Theory and Application,vol.9,no.8,pp.211-232,2016.

[37]XiaojunWuandSufangMeng."E-commercecustomer churn prediction based on improved SMOTE and AdaBoost."Inthe201613thInternationalConference onServiceSystemsandServiceManagement(ICSSSM), pp.1-5.IEEE,2016.