Effective implementation and governance of association rules in credit scoring and default management

Date
2022-04
Journal Title
Journal ISSN
Volume Title
Publisher
Stellenbosch : Stellenbosch University
Abstract
ENGLISH SUMMARY: Advancements in technology and an increase in the availability of alternative consumer data enabled an evolution in credit risk modelling. The result being traditional credit scoring and default management procedures becoming redundant. Encapsulating the alternative dimensions into one credit score is challenging, but machine learning techniques are well positioned to facilitate more accurate and modern credit scoring. To effectively leverage alternative data to increase the accuracy, validity and completeness of credit scores and reduce default rates, association rule learning is suggested. The extraction of associations between various alternative data and credit risk factors facilitates credit providers to assess credit risk more accurately but also enables early detection of possible credit defaults. Effectively implementing new technology, such as association rule learning, is a complex process. Therefore, to effectively implement association rules the features and characteristics relevant to implementation thereof needs to be considered against the life cycle of system development. Fundamental to the successful implementation of association rule Learning is sufficient quality data, therefore making proper data governance essential. The objective of this research was to provide a framework for credit providers to use when implementing association rules for credit scoring and default management. The framework is in the form of an application checklist that focuses on how to implement an unsupervised machine learning algorithm effectively, purposely considering the data governance issues faced by credit providers when implementing it for credit scoring and default management. To define association rule implementation guidelines for credit scoring and default management, this research provides an understanding of the evolution in the credit industry, the underlying data considerations and enabling technologies of big data, machine learning and data mining. The study further considers both the data governance and system development life cycle to identify applicable considerations relevant to implementing association rules for credit scoring and default management. Based on the investigation, an implementation checklist was designed to map the investigated implementation considerations with the system development life cycle, structured using the COBIT 2019 enablers. This model will provide guidance to data scientists and other users to assess their data and infrastructure requirements necessary and most appropriate for implementation.
AFRIKAANSE OPSOMMING: Tegnologiese vooruitgang en ’n toename in die beskikbaarheid van alternatiewe verbruikersdata het ’n evolusie in kredietrisikomodellering moontlik gemaak. Die resultaat was die vatbaarheid van tradisionele kredietgradering en wanbetaling-bestuursprosedures vir moontlike oortolligheid. Om die alternatiewe dimensies in een kredietgradering te omvat, is uitdagend, maar masjienleertegnieke is perfek geposisioneer om meer akkurate en moderne kredietgradering te fasiliteer. Die effektiewe gebruik van alternatiewe data om die akkuraatheid, geldigheid en volledigheid van kredietgradering te verhoog en verstekkoerse te verminder, word assosiasie-reel leerproses voorgestel. Deur die assosiasies tussen alternatiewe data- en kredietrisikofaktore te onttrek, word dit moontlik vir kredietverskaffers om kredietrisiko meer akkuraat te evalueer, en stel dit hulle ook in staat om moontlike krediet wanbetalers vroeg op te spoor. Die effektiewe implementering van nuwe tegnologie, soos die assosiasie-reelleer, is ’n komplekse proses wat min behaal word. Daarom moet die kenmerke en eienskappe wat relevant is vir die implementering van assosiasie-reels, geweeg te word teen die lewensiklus van stelselontwikkeling. Voldoende kwaliteit data is fundamenteel tot die suksesvolle implementering van assosiasie-reelleer en daarom is behoorlike databestuur noodsaaklik. Die doelwit van hierdie navorsing was om ‘n raamwerk aan kredietverskaffers te bied wanneer assosiasie-reëls geimplementeer word vir kredietgradering en wanbetalingsbestuur. Die raamwerk is in die vorm van ‘n toepassings kontrolelys wat fokus op hoe om effektiwilik masjienleeralgoritmes sonder toesig te implementeer en ter selfde tyd doelbewus te kyk na die databestuurkwessies wat kredietverskaffers ervaar wanneer dit geimplementeer word vir kredietgradering en wanbetalingsbestuur. Om die assosiasie-reel implementeringsriglyne vir kredietgradering en wanbetalingsbestuur te definieer, bied hierdie navorsing ’n begrip van die evolusie in die kredietbedryf, die onderliggende data-oorwegings en die bemagtigingtegnologiee van groot data, masjienleer en data-ontginning. Die studie oorweeg verder beide die databestuur en stelselontwikkelingslewensiklus om relevante oorwegings vir die implementering van assosiasie-reels vir kredietgradering en wanbetalingsbestuur te identifiseer. Op grond van die ondersoek is ’n implementeringskontrolelys ontwerp om die implementeringsoorwegings wat ondersoek is met die stelselontwikkelingslewensiklus te vergelyk, en met behulp van die COBIT 2019-instaatstellers te struktureer. Hierdie model sal datawetenskaplikes en ander gebruikers lei om hul vereistes vir data en infrastruktuur te evalueer vir nodigheid en geskiktheid ten einde dit te implementeer.
Description
Thesis (MCom)--Stellenbosch University, 2022.
Keywords
Citation