FIN-DM: a data mining process for the financial services
Date
2021-11-23
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Andmekaeve hõlmab reeglite kogumit, protsesse ja algoritme, mis võimaldavad ettevõtetel iga päev kogutud andmetest rakendatavaid teadmisi ammutades suurendada tulusid, vähendada kulusid, optimeerida tooteid ja kliendisuhteid ning saavutada teisi eesmärke. Andmekaeves ja -analüütikas on vaja hästi määratletud metoodikat ja protsesse. Saadaval on mitu andmekaeve ja -analüütika standardset protsessimudelit. Kõige märkimisväärsem ja laialdaselt kasutusele võetud standardmudel on CRISP-DM. Tegu on tegevusalast sõltumatu protsessimudeliga, mida kohandatakse sageli sektorite erinõuetega. CRISP-DMi tegevusalast lähtuvaid kohandusi on pakutud mitmes valdkonnas, kaasa arvatud meditsiini-, haridus-, tööstus-, tarkvaraarendus- ja logistikavaldkonnas. Seni pole aga mudelit kohandatud finantsteenuste sektoris, millel on omad valdkonnapõhised erinõuded.
Doktoritöös käsitletakse seda lünka finantsteenuste sektoripõhise andmekaeveprotsessi (FIN-DM) kavandamise, arendamise ja hindamise kaudu. Samuti uuritakse, kuidas kasutatakse andmekaeve standardprotsesse eri tegevussektorites ja finantsteenustes. Uurimise käigus tuvastati mitu tavapärase raamistiku kohandamise stsenaariumit. Lisaks ilmnes, et need meetodid ei keskendu piisavalt sellele, kuidas muuta andmekaevemudelid tarkvaratoodeteks, mida saab integreerida organisatsioonide IT-arhitektuuri ja äriprotsessi. Peamised finantsteenuste valdkonnas tuvastatud kohandamisstsenaariumid olid seotud andmekaeve tehnoloogiakesksete (skaleeritavus), ärikesksete (tegutsemisvõime) ja inimkesksete (diskrimineeriva mõju leevendus) aspektidega. Seejärel korraldati tegelikus finantsteenuste organisatsioonis juhtumiuuring, mis paljastas 18 tajutavat puudujääki CRISP- DMi protsessis.
Uuringu andmete ja tulemuste abil esitatakse doktoritöös finantsvaldkonnale kohandatud CRISP-DM nimega FIN-DM ehk finantssektori andmekaeve protsess (Financial Industry Process for Data Mining). FIN-DM laiendab CRISP-DMi nii, et see toetab privaatsust säilitavat andmekaevet, ohjab tehisintellekti eetilisi ohte, täidab riskijuhtimisnõudeid ja hõlmab kvaliteedi tagamist kui osa andmekaeve elutsüklis
Data mining is a set of rules, processes, and algorithms that allow companies to increase revenues, reduce costs, optimize products and customer relationships, and achieve other business goals, by extracting actionable insights from the data they collect on a day-to-day basis. Data mining and analytics projects require well-defined methodology and processes. Several standard process models for conducting data mining and analytics projects are available. Among them, the most notable and widely adopted standard model is CRISP-DM. It is industry-agnostic and often is adapted to meet sector-specific requirements. Industry- specific adaptations of CRISP-DM have been proposed across several domains, including healthcare, education, industrial and software engineering, logistics, etc. However, until now, there is no existing adaptation of CRISP-DM for the financial services industry, which has its own set of domain-specific requirements. This PhD Thesis addresses this gap by designing, developing, and evaluating a sector-specific data mining process for financial services (FIN-DM). The PhD thesis investigates how standard data mining processes are used across various industry sectors and in financial services. The examination identified number of adaptations scenarios of traditional frameworks. It also suggested that these approaches do not pay sufficient attention to turning data mining models into software products integrated into the organizations' IT architectures and business processes. In the financial services domain, the main discovered adaptation scenarios concerned technology-centric aspects (scalability), business-centric aspects (actionability), and human-centric aspects (mitigating discriminatory effects) of data mining. Next, an examination by means of a case study in the actual financial services organization revealed 18 perceived gaps in the CRISP-DM process. Using the data and results from these studies, the PhD thesis outlines an adaptation of CRISP-DM for the financial sector, named the Financial Industry Process for Data Mining (FIN-DM). FIN-DM extends CRISP-DM to support privacy-compliant data mining, to tackle AI ethics risks, to fulfill risk management requirements, and to embed quality assurance as part of the data mining life-cycle
Data mining is a set of rules, processes, and algorithms that allow companies to increase revenues, reduce costs, optimize products and customer relationships, and achieve other business goals, by extracting actionable insights from the data they collect on a day-to-day basis. Data mining and analytics projects require well-defined methodology and processes. Several standard process models for conducting data mining and analytics projects are available. Among them, the most notable and widely adopted standard model is CRISP-DM. It is industry-agnostic and often is adapted to meet sector-specific requirements. Industry- specific adaptations of CRISP-DM have been proposed across several domains, including healthcare, education, industrial and software engineering, logistics, etc. However, until now, there is no existing adaptation of CRISP-DM for the financial services industry, which has its own set of domain-specific requirements. This PhD Thesis addresses this gap by designing, developing, and evaluating a sector-specific data mining process for financial services (FIN-DM). The PhD thesis investigates how standard data mining processes are used across various industry sectors and in financial services. The examination identified number of adaptations scenarios of traditional frameworks. It also suggested that these approaches do not pay sufficient attention to turning data mining models into software products integrated into the organizations' IT architectures and business processes. In the financial services domain, the main discovered adaptation scenarios concerned technology-centric aspects (scalability), business-centric aspects (actionability), and human-centric aspects (mitigating discriminatory effects) of data mining. Next, an examination by means of a case study in the actual financial services organization revealed 18 perceived gaps in the CRISP-DM process. Using the data and results from these studies, the PhD thesis outlines an adaptation of CRISP-DM for the financial sector, named the Financial Industry Process for Data Mining (FIN-DM). FIN-DM extends CRISP-DM to support privacy-compliant data mining, to tackle AI ethics risks, to fulfill risk management requirements, and to embed quality assurance as part of the data mining life-cycle
Description
Keywords
financial services, data mining, data processing, data analysis