Reproduce “Machine Learning Models Evaluation and Feature Importance Analysis on NPL Dataset” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Rufael Fekadu, Anteneh Getachew, Yishak Tadele, Nuredin Ali, Israel Goytom
The paper evaluates machine learning models for predicting non-performing loans (NPLs) in an Ethiopian banking context, leveraging a proprietary dataset to identify key predictive features and effective imbalance treatment techniques. It shows that XGBoost achieves a 0.91 F1-score for NPLs when combined with K-Means SMOTE, and finds personal socio-economic factors to be more determinative of credit risk than collateral.
Key insights
- K-Means SMOTE is highly effective in improving model performance on imbalanced financial datasets, significantly outperforming CTGAN for generating synthetic samples.
- Personal and socio-economic characteristics such as applicant's age, total income, and years of employment are more predictive of loan default than collateral-related features in this context.
- Advanced ensemble models like XGBoost and Voting Classifiers demonstrate superior performance in predicting non-performing loans compared to traditional models.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.