ENHANCING CREDIT RISK PREDICTION AND INCLUSION THROUGH MACHINE LEARNING IN MICROFINANCE: A REVIEW by IRJET Journal

ENHANCING CREDIT RISK

PREDICTION AND INCLUSION THROUGH MACHINE LEARNING IN MICROFINANCE: A REVIEW

Namala Deekshitha ¹, Kareena Tunk ², Rathnavath Vyshnavi ³, Arjuman Subhani ⁴

1Stanley College of engineering and technology for women, India

2Stanley College of engineering and technology for women, India

³Stanley College of engineering and technology for women, India

Asst. Professor, Dept. of AI&DS and CME Engineering, Stanley College of engineering and technology for women, India

Abstract - By giving underprivileged groups access to credit and other financial products, microfinance institutions (MFIs) are essential in the provision of financial services. Microfinance has changed as a result of the incorporation of machine learning (ML), deep learning (DL) which has improved financial decision-making, loan default prediction, and credit risk assessment. With an emphasis on enhancing credit scoring models, streamlining loan approval procedures, and reducing financial risks, this study investigates the use of diverse machine learning approaches in microfinance. MFIs can improve portfolio management, lower default rates, and advance financial inclusion by utilising predictive algorithms like decision trees, random forests, and neural networks. The study highlights upcoming innovation opportunities and gives a summary of recent developments in machine learning within the microfinance industry.

Key Words: MicrofinanceInstitutions(MFIs),Machinelearning(ML),Deeplearning(DL),Financialdecision-making,Loan defaultprediction,Creditriskassessment,Decisiontrees,Randomforests.

1.INTRODUCTION

Microfinanceinstitutions(MFIs)havebeenessentialinhelpingsmallbusinessesandindividualswithoutaccesstotraditional bankingsystemsbyofferingfinancialservices.MFIssupportfinancialinclusion,especiallyinunderservedareas,byproviding credit, savings, and insurance products. But historically, MFIs' growth and sustainability have been hampered by their particular set of problems, which include high default rates, operational inefficiencies, and a lack of information about borrowers.

Machinelearning(ML)hasbecomea potenttoolinrecentyearstotackletheseissues.MFIs canbetterassesscredit risk, anticipateloandefaults,processvastvolumesofdataefficiently,andmakebetterdecisionsregardingloanapprovalsthanksto machine learning techniques. Machine learning can greatly improve the speed and accuracy of financial assessments in microfinancebyutilisingpredictivemodelslikedecisiontrees,randomforests,andneuralnetworks.Thiswilllowerdefault ratesandimprovefinancialperformance.

This review of the literature looks at how machine learning methods are used in the microfinance industry. It highlights importantstudiesthathaveusedmachinelearning(ML)topredictloanperformance,managerisk,andscorecredit.Alongwith highlightingtrends,obstacles,andareasforfurtherstudy,thereviewprovidesinsightsintohowmachinelearningcankeep revolutionisingthemicrofinancesector.Thispaperattemptstogiveathoroughgraspoftherelationshipbetweenmicrofinance and machine learning by synthesising recent research, demonstrating the potential of these technologies to enhance sustainabilityandfinancialinclusion.

Thisreviewisorganizedasfollows:Section2,givestheoverviewofmicrofinanceandmachinelearning.Section3,providesthe listofprominentresearchworkthatwasdoneanddifferenttechniquesthatareemployed.Section4,describessomeofthe mostprominentfutureresearchlines.Lastly,conclusionsareprovidedintheSection5.

2. OVERVIEW OF MICROFINANCE AND MACHINE LEARNING

Asubsetoffinancialservicesknownasmicrofinanceprovidessmallloanstolow-incomeindividualswhomightnototherwisebe abletoobtainorqualifyfortraditionalfinancing[1][2].Economicdevelopment,financialinclusion,andpovertyreductionhave allbeenshowntobepossiblewithmicrofinance[3][4][5].Manymicrofinanceinstitutions(MFIs)havefacedsustainability challengesdespitetheirdemonstratedpotential,mainlyasaresultofrisingloandefaultrates[6][7]

e-ISSN: 2395-0056 p-ISSN: 2395-0072 Volume: 12 Issue: 03 | Mar 2025 www.irjet.net

2.1 Challenges in Microfinance

MFIsdealwiththefollowingproblems.Tobegin,existingtechniquessuchasbinaryclassificationorcreditscore-basedmethods mostlyfocusonofflinelearning,assumingthattrustworthyhomogeneousdataisgenerallyavailable.Therefore,suchlearningor credit-scoringapproachescannotbeimmediatelyappliedtomicrofinancewithoutpriorloanhistoriesorsuitablefinancial systemstoputthemupaccurately[8][9].Additionally,duetoalackofsuitabletechniques,someapplicantsfinditdifficultto provideadequateinformationtoaccuratelyevaluatetheircreditratingsanddefaultlikelihood[10]

Second,theexistenceofnumerouspeopleandareasmakesitmorechallengingtodistributemicrofinanceresourcesinaway thatbalancesdifferentfairness/inclusionaims.MFIshavemainlyreliedonthejudgementofloanofficersduetothelackof methodologiesthatcansystematicallybalancetherisks,fairness,andmultifacetedobjectivesofmicrofinance.Asaresult, decisionshaveoccasionallyallowedPortfolioatRisk(PAR)toexceedalevelnecessarytosustainmicrofinanceoperations.Due tothelackofproceduresthatcansystematicallybalancetherisks,fairness,andmultifariousaimsofmicrofinance,MFIshave primarilyreliedonthejudgementofloanofficers.DecisionshavethereforeoccasionallypermittedPortfolioatRisk(PAR)to surpassthethresholdrequiredtomaintainmicrofinanceoperations.

2.2 Role of Machine Learning:

Potentialanswerstotheseproblemscanbefoundinmachinelearning(ML).Insituationswherethereisalackoffinancial data,machinelearning(ML)canenhancecreditriskassessmentbyutilisinglargedatasetsandcreatingpredictivemodels.Loan defaultprobabilitiescanbepredictedusingsupervisedlearningalgorithmslikedecisiontreesandneuralnetworks,andMFIs can better segment their customer bases using clustering techniques even in environments with heterogeneous data. Furthermore,bybalancingriskandinclusiongoals,fairness-awaremachinelearningmodelscanhelpmakedecisionsloan distributionthataremoreequitable.

MFIscanimprovetheirdecision-makingprocedures,lessentheirdependencyonhumanjudgement,andeventuallyincrease thesustainabilityoftheiroperationsbyimplementingmachinelearningapproaches.

3.COMPARISION OF RESEARCH WORKS

Thissectioncomparesseveralstudiesthatexaminehowmachinelearninganddeeplearningapproachesareappliedinthe microfinance industry. The comparison highlights important elements like the datasets used, the algorithms used, the performanceoutcomes,andthedifficultiesfaced.

Table -1: Comparisonofresearchwork

Ref.no Title Year Dataset Used Algorithm Implemented Performance Result (Accuracy) Challenges Encountered

[11] AMachineLearning Approach for Micro-CreditScoring

[12] Predicting Loan Defaults using MachineLearning

[13] An Exploration of AlternativeFeatures in Microfinance Loan Default Prediction Models

2021 Micro-lending data from developing regions, includes demographics (age, occupation, location)

2019 Nigerian credit bureau, SMS, and appdata

2020 Combined dataset with alternative and traditional credit data

Decision Trees, SVM, Random Forest 85% accuracy with Random Forest Lack of formal credit history for borrowers; imbalanceddataset

Logistic Regression, RandomForest, XGBoost,FCNN

XGBoost:~85% accuracy Model complexity, interpretability challenges with neuralnetworks

XGBoost: 87% accuracy Data privacy and ethicalchallenges

International Research

Volume: 12 Issue: 03 | Mar 2025 www.irjet.net

[14] Predicting the Performance of Rural Banks in Ghana Using MachineLearning

[15] Rural Micro Credit Assessment Using MachineLearningin Peru

[16] Improving the Management of MFIs Using Credit ScoringModels

[17] A Learning and Control Perspective forMicrofinance

[18] A Deep Learning Approach to Risk Management for IslamicMicrofinance

[19] Exploring the Influence of Microfinance on Entrepreneurship

[20] Credit Scoring in MicrofinanceUsing Non-Traditional Data

[21] Neural Network Credit Scoring Models

[22] A Deep Learning Based Online Credit Scoring Model for P2PLending

[23] Fuzzy Logic Approach Applied to Credit Scoring forMicrofinancein

of Engineering and Technology (IRJET) e-ISSN: 2395-0056 p-ISSN: 2395-0072

2020 Rural bank performance data, includes economic and customer financialdata

2018 Peruvian microfinance data, includes loan amounts,repayment history

2019 MFI customer data, including loan and repaymentbehavior

2021 Notspecifieddueto theoreticalapproach

2021 Islamic finance data, including income, expenses, and debt ratios

2020 Microfinance and entrepreneurship data

2020 Non-traditional (unstructured)data from digital interactions

2018 Credit card and loan datasets, including credit limit, balances, and repaymenthistory

2020 P2P lending data, including borrower history, credit scores, and transactiondata

2018 Moroccan microfinance dataset with 78 fuzzy rules applied

FCNN, Logistic Regression 90% accuracy with Neural Networks Limited data availability; challenges in collecting standardizeddata

Bayesian Networks 75%accuracy Data sparsity and quality issues due to inconsistent records

Logistic Regression, DecisionTrees 80% accuracy with Decision Trees Model interpretabilityand transparency in creditscoring

Control theorybased machine learning Notspecified Adapting control theory to financial models remains challenging

LSTM 85% accuracy withLSTM

SVM, Decision Trees,FCNN 70% accuracy withSVM

Data labeling and quality in Islamic financecontexts

Data consistency and integrating diverse data sources; limited entrepreneurship data

RandomForest, Logistic Regression 75% accuracy with Random Forest Integrating unstructured data sources and addressing quality control

FCNN 85%accuracy

FCNN,LSTM 90% accuracy withLSTM

Model training time and computational resourceintensity

Model overfitting and challenges in real-time predictions

FuzzyLogic Notspecified Addressing information asymmetry and lack of precise

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 p-ISSN: 2395-0072

Volume: 12 Issue: 03 | Mar 2025 www.irjet.net

[24] Loan Risk Prediction Using Machine Learning Algorithms: The Case of Ethiopia’s Micro-Finance Institutions

[25] Rural Micro Credit Assessment Using MachineLearningin Peruvian micro financeinstitution

2023 Ethiopian microfinance data (18,308 records on loanstatus)

Random Forest, Decision Tree, XGBoost,MLP

XGBoost: 98% accuracy Dataqualityissues, imbalanced data, and complexity in handling diverse borrower demographics

2021 Dataset with 15,015 clients’ data, focused onruralPeru

ArtificialNeural Network(ANN), Logistic Regression, RandomForest, SVM, Decision Tree,KNN

4.FUTURE SCOPE

ANN: 93.72% accuracy High variability in customer data; ensuring data quality in remote rural regions; addressing model biases in credit scoring due to limitedinformation

Whilemachinelearninghassignificantlyimprovedcreditriskassessmentandloandefaultpredictioninmicrofinance,there areseveralareasforfurtherdevelopment:

• ImprovingDataAccessandQuality:It'scriticaltoimprovedataaccessibilityandquality,particularlyinunderserved areas.Modelperformanceandtrainingcanbeenhancedbymethodssuchasdataaugmentationandopenfinancialdatasets.

• FairnessandBias:Tomakesurethatfinancialinclusioninitiativesdonotinadvertentlyintroducebias,especiallyin diverse,low-incomepopulations,futuremodelsshouldconcentrateonfairness-awarealgorithms.

• ModelInterpretability:BymakingMLpredictionsmoreunderstandableandtrustworthythroughexplainableAI(XAI), MFIswillbeabletoimplementthemmoresmoothly.

• Real-TimeCreditScoring:Bycreatingmobile-based,real-timecreditscoringsystems,loanapprovalprocedurescanbe enhancedinareaswithinadequateinfrastructure,allowingforpromptdecision-making.

• Low-ResourceModelAdaptation:Bydevelopinglightweight,inexpensivemodels,machinelearningwillbecomemore accessibletoMFIsthatworkinresource-constrainedsettings,improvingscalability.

• BlockchainIntegration:Bycombiningblockchaintechnologywithmachinelearning,microfinancetransactionscan becomemoretransparentandsecure,reducingtheriskoffraudandguaranteeingdataintegrity.

• Personalised Financial Products: To increase customer satisfaction and repayment rates, future research can concentrateoncreatingfinancialproductsandloansthatarespecificallysuitedtoborrowerbehaviour.

Futurestudiesandinnovationscanfurtheradvancetheuseofmachinelearninginmicrofinancebytacklingtheseissues,which willincreaseoperationaleffectivenessandfinancialinclusion.Thefutureofmicrofinancewillbesignificantlyshapedbymachine learningtechnologiesastheydevelop,contributingtothedevelopmentofamoresustainableandinclusivefinancialsystem.

5.CONCLUSION

Theexpandingroleofdeeplearningandmachinelearninginrevolutionisingthemicrofinanceindustryhasbeenexaminedin this review. Microfinance institutions (MFIs) can improve their operational efficiency and decision-making processes by utilising these technologies to improve financial inclusion, loan default prediction, and credit risk assessment. MFIs can optimiseloanapprovals,lowerdefaultrates,andassessborrowerriskmoreaccuratelybyutilisingalgorithmssuchasdecision trees,randomforests,andneuralnetworks.

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 p-ISSN: 2395-0072

Volume: 12 Issue: 03 | Mar 2025 www.irjet.net

However,issueslikefairness,interpretabilityofthemodel,anddataqualitystillexistandcallformorestudyandadvancement. To fully utilise machine learning in microfinance, these problems must be resolved, real-time credit scoring must be incorporated,andmodelsmustbemodifiedforlow-resourcesettings.

Theuseofmachinelearninginmicrofinance,particularlyinunderprivilegedareas,willbeessentialtopromotinggreater financial inclusionand sustainability asthe technology develops.Byadoptingthese technologies,MFIs can enhance their customerserviceandhelpreducepovertybyprovidingbetteraccesstofinancialservices.

ACKNOWLEDGEMENT

"We express our heartfelt gratitude to Ms.Arjuman Subhani, Assistant Professor, Stanley College Of Engineering and TechnologyforWomen,fortheirinvaluableguidance,continuousencouragement,andinsightfulfeedbackthroughout the course of this research. Their expertise and support were instrumental in shaping this work, and weare deeply thankful fortheirmentorship."

REFERENCES

[1] J. Morduch and B. Armendariz, The Economics of Microfinance. Cambridge, Mass: MIT Press, 2005. Available: https://nyuscholars.nyu.edu/en/publications/the-economics-of-microfinance

[2] M. RashidandKamanza,“CausesofDefaultonMicro-CreditamongWomenMicro-EntrepreneursinKenya.ACaseStudy ofWomenEnterpriseDevelopmentFund(Wedf)MsambweniConstituency,”IOSRJournalofEconomicsandFinance,vol.3, no.6,pp.32–47,2014.

[3] R. Mersland and R. Ø. Strøm, “Microfinance Mission Drift?,” World Development, vol. 38, no. 1, pp. 28–36, Jan. 2010, doi:20https://doi.org/10.1016/j.worlddev.2009.05.006.

[4] N.HermesandM.Hudon,“DeterminantsofthePerformanceofMicrofinanceInstitutions:ASystematicReview,”University ofGroningenresearchdatabase(UniversityofGroningen/CentreforInformationTechnology),pp.297–330,Mar.2019, doi:https://doi.org/10.1002/9781119565178.ch10.

[5] C.MilanaandA.Ashta,“Microfinanceandfinancialinclusion:Challengesandopportunities,”StrategicChange,vol.29,no.3, pp.257–266,May2020,doi:https://doi.org/10.1002/jsc.2339.

[6] N.NawaiandM.N.M.Shariff,“FactorsAffectingRepaymentPerformanceinMicrofinanceProgramsinMalaysia,”Procedia -SocialandBehavioralSciences,vol.62,pp.806–811,Oct.2012,doi:https://doi.org/10.1016/j.sbspro.2012.09.136.

[7] A.Addae-Korankye,“CausesandControlofLoanDefault/DelinquencyinMicrofinanceInstitutionsinGhana,”American International Journal of Contemporary Research, vol. 4, no. 12, 2014, Available: https://www.aijcrnet.com/journals/Vol_4_No_12_December_2014/5.pdf

[8] J.N.NjagiandC.Njoka,“MicrofinanceReformsandFinancialInclusioninKenya,”InternationalJournalofCurrentAspects inFinance,BankingandAccounting,vol.3,no.1,pp.54–72,Aug.2021,doi:https://doi.org/10.35942/ijcfa.v3i1.181.

[9] ArwaAbubakerAlamoudiandAnwar,“AvailableFinancingResourcesForIslamicMicrofinanceInstitutionsToAlleviate Poverty- Cash Waqf Approach,” Journal of Islamic Finance, vol. 10, pp. 076–084076–084, Oct. 2022, doi: https://doi.org/10.31436/jif.v10i.529.

[10] H.MaulanaandK.Umam,“IdentifyingFinancialExclusionandIslamicMicrofinanceasAnAlternativetoEnhanceFinancial Inclusion,” International Journal of Islamic Business and Economics (IJIBEC), p. 99, Jan. 2018, doi: https://doi.org/10.28918/ijibec.v1i2.1004.

[11] A.Ampountolas,T.NyarkoNde,P.Date,andC.Constantinescu,“AMachineLearningApproachforMicro-CreditScoring,” Risks,vol.9,no.3,p.50,Mar.2021,doi:https://doi.org/10.3390/risks9030050.

[12] A.Bhagat,2018.Available:https://scholarworks.calstate.edu/downloads/8k71nk666

[13] D.StoneandM.Britz,“AnExplorationofAlternativeFeaturesinMicro-FinanceLoanDefaultPredictionModels,”2020.

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 p-ISSN: 2395-0072

Volume: 12 Issue: 03 | Mar 2025 www.irjet.net

[14] E.Awoin,P.Appiahene,F.Gyasi,andA.Sabtiwu,“PredictingthePerformanceofRuralBanksinGhanaUsingMachine Learning Approach,” Advances in Fuzzy Systems, vol. 2020, pp. 1–7, Feb. 2020, doi: https://doi.org/10.1155/2020/8028019.

[15] E.Morales,J.Aguirre,B.Ramos,andD.Sanchez,“CreditRiskAnalysisModelinMicrofinanceInstitutionsinPeruThrough theuseofBayesianNetworks.”

[16] M.-D.Cubiles-De-La-Vega,A.Blanco-Oliver,R.Pino-Mejías,andJ.Lara-Rubio,“Improvingthemanagementofmicrofinance institutionsbyusingcreditscoringmodelsbasedonStatisticalLearningtechniques,”ExpertSystemswithApplications, vol.40,no.17,pp.6910–6917,Dec.2013,doi:https://doi.org/10.1016/j.eswa.2013.06.031.

[17] X.Dengetal.,“ALearningandControlPerspectiveforMicrofinanceRegionalICTCenterofExcellenceBldgPlotNoA8, KigaliSpecialEconomicZonePhaseII,Rwanda,”ProceedingsofMachineLearningResearch,vol.211,pp.1–13,2023.

[18] K.KatterbauerandP.Moschetta,“AdeeplearningapproachtoriskmanagementmodelingforIslamicmicrofinance,”

EuropeanJournalofIslamicFinance,pp.2421–2172,doi:https://doi.org/10.13135/2421-2172/6202-Published.

[19] Z.Malik,N.Ahmad,andW.Ahmed,“ExploringtheInfluenceofMicrofinanceonEntrepreneurshipusingmachinelearning techniques Exploring the Influence of Microfinance on Entrepreneurship using machine…,” Journal of Information TechnologyManagement,vol.15,pp.139–156,2023,doi:https://doi.org/10.22059/jitm.2023.95250.

[20] S. Ruiz, P. Gomes, L. Rodrigues, and J. Gama, “Credit Scoring in Microfinance Using Non-traditional Data,” doi: https://doi.org/10.1007/978-3-319-65340-2.

[21] D.West,“NeuralNetworkCreditScoringModels,”Computers&OperationsResearch,vol.27,pp.1131–1152,2000.

[22] Z.Zhang,K.Niu,andY.Liu,“ADeepLearningBasedOnlineCreditScoringModelforP2PLending,”IEEEAccess,vol.8,pp. 177307–177317,2020,doi:https://doi.org/10.1109/access.2020.3027337.

[23] G.Bennouna,“ScienceDirectScienceDirectTheFirstInternationalConferenceOnIntelligentComputinginDataSciences Fuzzylogicapproachappliedtocred-NCandpeer-reviewunderresponsibilityofInternationalNeuralNetworkSociety Morocco Regional Chapter,” Procedia Computer Science, vol. 127, pp. 274–283, 2018, doi: https://doi.org/10.1016/j.procs.2018.01.123.

[24] D.Bizuwork,H.Advisor,andG.Semeon,“LoanRiskPredictionUsingMachineLearningAlgorithms:-TheCaseofEthiopia’s Micro-FinanceInstitution’s,”2019.

[25] I.Henry,Condori-Alejo,M.RomilioAceituno-Rojo,andG.Alzamora,“ScienceDirectRuralMicroCreditAssessmentusing MachineLearninginaPeruvianmicrofinanceinstitution,”ProcediaComputerScience,vol.187,pp.408–413,2021,doi: https://doi.org/10.1016/j.procs.2021.04.117.