Analiza tveganj pri spletnih posojilih : delo diplomskega seminarja

Zukanovič, Jure (author), Orbanić, Alen (mentor)

Rdeča nit tega dela diplomskega seminarja je predstavitev in primerjava različnih metod strojnega učenja za identificiranje "dobrega izposojevalca", na podlagi podatkov priljubljene spletne platforme za posojanje denarja Lending Club (kratica LC). V delu so predstavljeni rezultati ugotovitev članka z naslovom Risk assessment in social lending via random forests, avtorjev Milada Malekipirbazarija in Vurala Aksakallija, ki sta z uporabo programa WEKA prikazala, da je matoda naključnih gozdov pri identificiranju "dobrega izposojevalca" boljša od točk FICO in ocene s strani LC. To sta metodi, ki ju za identificiranje "dobrega izposojevalca" dandanes uporabljajo agencije, ki se ukvarjajo z računanjem kreditne ocene posameznika za spletne posojilnice. V zaključku dela je predstavljena rekonstrukcija in potrditev navedb članka na podlagi rezultatov, ki sem jih dobil, ko sem simulacije v programu WEKA zagnal še sam. Metode strojnega učenja, ki so predstavljene v tem delu diplomskega seminarja, so: metoda naključnih gozdov, logistična regresija, metoda podpornih vektorjev in metoda k-najbližjih sosedov.

• Language: Slovenian
• Keywords: finančna matematika, posojanje denarja prek spleta, ocena tveganja, spletna posojilnica, posojevalec, izposojevalec, strojno učenje, naključni gozd, logistična regresija, metoda podpornih vektorjev, metoda k-najbližjih sosedov
• Work type: Final seminar paper
• Typology: 2.11 - Undergraduate Thesis
• Organization: FMF - Faculty of Mathematics and Physics
• Place of publishing: Ljubljana
• Publisher: [J. Zukanovič]
• Year: 2017
• Number of pages: 28 str.
• PID: 20.500.12556/RUL-106212
• UDC: 512
• COBISS.SI-ID: 18553945
• Publication date in RUL: 12.02.2019

ZUKANOVIČ, Jure, 2017, Analiza tveganj pri spletnih posojilih : delo diplomskega seminarja [online]. Bachelor’s thesis. Ljubljana : J. Zukanovič. [Accessed 2 April 2025]. Retrieved from: https://repozitorij.uni-lj.si/IzpisGradiva.php?lang=eng&id=106212

Secondary language

• Language: English

Abstract

The cohesive thread of the following seminar thesis is the presentation and comparison of different methods of machine learning to identify a "good borrower", according to the popular online lending platform Lending Club (acronym LC). The paper presents the results of the findings of an article entitled Risk assessment in social lending via random forests by Milad Malekipirbazari and Vural Aksakalli, who, with the help of WEKA program, showed that the random forests method in identifying a "good borrower" is better than FICO score and LC grades. These methods for identifying a "good borrower" are used by today's agencies to identify the credit rating of an individual for social lending. In the conclusion of the thesis's seminar, a reconstruction and confirmation of results of the article is presented, based on results I have obtained after running the simulations in WEKA program first-hand. Machine learning methods presented in this thesis's seminar are: random forests, logistic regression, support vector machines, and k-nearest neighbours method.

• Keywords: mathematics, peer-to-peer lending, P2P lending, risk assessment, social lending platform, lender, borrower, machine learning, random forests, logistic regression, support vector machines, k-nearest neighbors