University of Twente Student Theses


Improving a risk-based revision model for SME loans

Otten, T. (2022) Improving a risk-based revision model for SME loans.

[img] PDF
Abstract:In this research, we study how we can improve the risk-based revision model of a Dutch bank that signals the need for measures that should prevent clients from going into default based on the current financial situation of clients. We test various machine learning models, namely logistic regression, decision trees and random forests, to obtain a new improved classification model. Most of the clients of the bank are in a healthy financial situation and do not need measures to prevent a default. Therefore, we implement imbalanced dataset techniques such as alternative cut-off strategies and the synthetic minority oversampling technique (SMOTE) to boost the performance of the models. Also, we apply hyperparameter tuning by implementing a grid search. Higher precision is preferred over lower recall by management in this study. Therefore, we measure model performance with the F0.33 score. We found that by using a random forest model and applying an alternative cut-off point, we could improve the F0.33 score from 7 percent to 71 percent, indicating that the new model can be seen as an improvement over the old model. Underlying here is that precision increased from 6 percent to 84 percent but the recall performance decreased from 80 percent to 30 percent. Besides suggesting a configuration for the new model, we show what configurations are possible in terms of different precision versus recall trade-offs to provide insight to management on what performance levels can be achieved. Keywords: Forbearance measures, risk-based revisions, logistic regression, decision trees, random forests, imbalanced dataset techniques, alternative cut-off, SMOTE, grid search
Item Type:Essay (Master)
Faculty:BMS: Behavioural, Management and Social Sciences
Programme:Industrial Engineering and Management MSc (60029)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page