Bitte benutzen Sie diese Kennung, um auf die Ressource zu verweisen:
https://doi.org/10.21256/zhaw-24573| Publikationstyp: | Thesis: Bachelor |
| Titel: | Using deep learning and explainable AI to predict and explain loan defaults |
| Autor/-in: | Sulejmani, Sakip |
| Betreuer/-in / Gutachter/-in: | Fazlija, Bledar |
| DOI: | 10.21256/zhaw-24573 |
| Umfang: | 47 |
| Erscheinungsdatum: | 2021 |
| Verlag / Hrsg. Institution: | ZHAW Zürcher Hochschule für Angewandte Wissenschaften |
| Verlag / Hrsg. Institution: | Winterthur |
| Sprache: | Englisch |
| Fachgebiet (DDC): | 006: Spezielle Computerverfahren 332: Finanzwirtschaft |
| Zusammenfassung: | The use of machine learning in finance is increasing, and while deep learning models are becoming the state of the art to make predictions, the difficulty of interpreting them is a drawback. This is especially so in finance, where each result that a model outputs must be explainable and justifiable. In recent years, novel explainable AI methods have been researched and developed to explain deep learning models and their decisions. The aim of this bachelor thesis was to analyze a use case in credit scoring, specifically in loan defaulting, with deep learning and explainable AI. It also aimed to show that deep learning can be used to predict loan defaults in finance, that explainable AI methods offer insights for interpreting the black box’s internal decisions, and furthermore, that it is possible to improve models with insights from explainable AI. A peer-to-peer loan dataset from Bondora with 164,547 instances and 112 features was analyzed, pre-processed, and prepared for deep learning. Multiple neural networks with different parameters were fitted and evaluated to find the best hyperparameters for loan default predicting with the dataset. A post hoc analysis with SHAP was applied to the best model to retrieve insights from it. These insights were then used to explain the model’s decisions and to adjust it. The results show that the model has an AUC of 0.72 and can therefore differentiate between a defaulted and a not defaulted loan with a probability of 72%. In addition, a recall of 0.88 was reached, meaning the model predicts 88% of defaulted loans correctly. Furthermore, the insights gained from explainable AI enabled the creation of a second, adjusted model that reached equally good performance with only half of the features. Moreover, the explainable AI insights were used to determine and analyze the fifteen features which influence the model the most. The three most influential were debt-to-income, applied amount and loan duration. Additionally, two loan instances from the dataset were analyzed in detail with SHAP. In conclusion, using deep learning and explainable AI we were able to predict loan defaults, and interpret as well as explain the model’s decisions. Moreover, the explainable AI insights could be used to adjust and improve the model. A complete use case in credit scoring is shown in this thesis, highlighting that deep learning and explainable AI can be used in finance. However, the gained insights from the explainable AI methods were very specific to the used dataset and therefore further research with different datasets would be interesting. |
| URI: | https://digitalcollection.zhaw.ch/handle/11475/24573 |
| Lizenz (gemäss Verlagsvertrag): | CC BY-NC-ND 4.0: Namensnennung - Nicht kommerziell - Keine Bearbeitungen 4.0 International |
| Departement: | School of Management and Law |
| Enthalten in den Sammlungen: | BSc Wirtschaftsinformatik |
Dateien zu dieser Ressource:
| Datei | Beschreibung | Größe | Format | |
|---|---|---|---|---|
| BA_Sakip_Sulejmani.pdf | 5.85 MB | Adobe PDF |  Öffnen/Anzeigen |
Zur Langanzeige
Sulejmani, S. (2021). Using deep learning and explainable AI to predict and explain loan defaults [Bachelor’s thesis, ZHAW Zürcher Hochschule für Angewandte Wissenschaften]. https://doi.org/10.21256/zhaw-24573
Sulejmani, S. (2021) Using deep learning and explainable AI to predict and explain loan defaults. Bachelor’s thesis. ZHAW Zürcher Hochschule für Angewandte Wissenschaften. Available at: https://doi.org/10.21256/zhaw-24573.
S. Sulejmani, “Using deep learning and explainable AI to predict and explain loan defaults,” Bachelor’s thesis, ZHAW Zürcher Hochschule für Angewandte Wissenschaften, Winterthur, 2021. doi: 10.21256/zhaw-24573.
SULEJMANI, Sakip, 2021. Using deep learning and explainable AI to predict and explain loan defaults. Bachelor’s thesis. Winterthur: ZHAW Zürcher Hochschule für Angewandte Wissenschaften
Sulejmani, Sakip. 2021. “Using Deep Learning and Explainable AI to Predict and Explain Loan Defaults.” Bachelor’s thesis, Winterthur: ZHAW Zürcher Hochschule für Angewandte Wissenschaften. https://doi.org/10.21256/zhaw-24573.
Sulejmani, Sakip. Using Deep Learning and Explainable AI to Predict and Explain Loan Defaults. ZHAW Zürcher Hochschule für Angewandte Wissenschaften, 2021, https://doi.org/10.21256/zhaw-24573.
Alle Ressourcen in diesem Repository sind urheberrechtlich geschützt, soweit nicht anderweitig angezeigt.