prikaz prve stranice dokumenta Procjena kreditnog rizika malih i srednjih poduzeća probranim metodama strojnog i dubokog učenja
  Download
PDF 31.04 MB
master's thesis
Procjena kreditnog rizika malih i srednjih poduzeća probranim metodama strojnog i dubokog učenja
2022. urn:nbn:hr:126:197359
Krizmanić, Antonio
Josip Juraj Strossmayer University of Osijek
Department of Mathematics

Cite this document

Krizmanić, A. (2022). Procjena kreditnog rizika malih i srednjih poduzeća probranim metodama strojnog i dubokog učenja (Master's thesis). Osijek: Josip Juraj Strossmayer University of Osijek, Department of Mathematics. Retrieved from https://urn.nsk.hr/urn:nbn:hr:126:197359

Krizmanić, Antonio. "Procjena kreditnog rizika malih i srednjih poduzeća probranim metodama strojnog i dubokog učenja." Master's thesis, Josip Juraj Strossmayer University of Osijek, Department of Mathematics, 2022. https://urn.nsk.hr/urn:nbn:hr:126:197359

Krizmanić, Antonio. "Procjena kreditnog rizika malih i srednjih poduzeća probranim metodama strojnog i dubokog učenja." Master's thesis, Josip Juraj Strossmayer University of Osijek, Department of Mathematics, 2022. https://urn.nsk.hr/urn:nbn:hr:126:197359

Krizmanić, A. (2022). 'Procjena kreditnog rizika malih i srednjih poduzeća probranim metodama strojnog i dubokog učenja', Master's thesis, Josip Juraj Strossmayer University of Osijek, Department of Mathematics, accessed 11 November 2024, https://urn.nsk.hr/urn:nbn:hr:126:197359

Krizmanić A. Procjena kreditnog rizika malih i srednjih poduzeća probranim metodama strojnog i dubokog učenja [Master's thesis]. Osijek: Josip Juraj Strossmayer University of Osijek, Department of Mathematics; 2022 [cited 2024 November 11] Available at: https://urn.nsk.hr/urn:nbn:hr:126:197359

A. Krizmanić, "Procjena kreditnog rizika malih i srednjih poduzeća probranim metodama strojnog i dubokog učenja", Master's thesis, Josip Juraj Strossmayer University of Osijek, Department of Mathematics, Osijek, 2022. Available at: https://urn.nsk.hr/urn:nbn:hr:126:197359

Please login to the repository to save this object to your list.
Metadata
TitleProcjena kreditnog rizika malih i srednjih poduzeća probranim metodama strojnog i dubokog učenja
Title (english)Credit Risk Assessment for SMEs Using Machine and Deep Learning Methods
AuthorAntonio Krizmanić
MentorNataša Šarlija (mentor)
MentorSlobodan Jelić (komentor)
Committee memberDomagoj Matijević (predsjednik povjerenstva)
Committee memberNataša Šarlija (član povjerenstva)
Committee memberDomagoj Ševerdija (član povjerenstva)
GranterJosip Juraj Strossmayer University of Osijek
Department of Mathematics
Osijek
Defense date and country2022-09-27, Croatia
Scientific / art field,
discipline and subdiscipline		NATURAL SCIENCES
Mathematics
Probability Theory and Statistics
Abstract
Cilj istraživanja bilo je usporediti performanse CART metode, metode slučajne šume, metode ekstremnog podizanja gradijenta te metode višeslojnog perceptrona u procjeni kreditnoga rizika za iz javnih baza podataka uzorkovana mala i srednja poduzeća s područja Republike Hrvatske. Istraživanje je provedena u tri faze: između skupova najboljih 1% predstavnika svake metode te izmedu pojedinačnih najboljih predstavnika metoda prije i poslije selekcije relevantnih pokazatelja SFFS algoritmom. Hijerarhija kvalitete predstavnika unutar četiriju familija metoda utvrđena je algoritmom pretrage mreže hiperparametara. Zabilježena je dominacija ansambla modela nad pojedinačnim modelima pri čemu je, generalno govoreći, XGBoost postigao najbolje rezultate, praćen slučajnom šumom, perceptronom te, na posljetku, klasifikacijskim stablom. Istraživanje valja proširiti dodatnim metodama (LR, SVM), složenijim metodama dubokog učenja (CNN, RNN, TabNet) te ga obogatiti većim skupom jedinki opisanih većim brojem pokazatelja.
Abstract (english)
The objective of the research was to compare the performances of the CART, random forest, extreme gradient boosting and multilayer perceptron methods in credit risk assessment for small and medium-sized enterprises sampled from public databases in the Republic of Croatia. We conducted the research in three phases: between the 1% of the best representatives of each method and between the individual best representatives of the methods before and after selecting correspondent relevant features using the SFFS algorithm. The hierarchy of the quality of the representatives within the four families of methods was established by the hyperparameter grid search algorithm. We noted a trend of ensemble methods dominating individual models, with XGBoost achieving the best results in general, followed by random forest, perceptron and, finally, the classification tree. The research should be expanded with additional methods (LR, SVM), more complex deep learning methods (CNN, RNN, TabNet) and enriched with a greater set of individuals described by a greater number of indicators.
Keywords
Keywords (english)
Languagecroatian
URN:NBNurn:nbn:hr:126:197359
Study programmeTitle: Mathematics; specializations in: Financial and Statistical Mathematics, Mathematics and Computer Science, Industrial and Applied Mathematics
Course: Financial and Statistical Mathematics
Study programme type: university
Study level: graduate
Academic / professional title: magistar/magistra matematike (magistar/magistra matematike)
Type of resourceText
File originBorn digital
Access conditionsOpen access
Terms of useLicence In copyright icon
Created on2022-09-29 06:42:22