[Paper Review] Predicting Credit Risk for Unsecured Lending: A Machine Learning Approach

Kevin Naik|arXiv (Cornell University)|Oct 5, 2021

Financial Distress and Bankruptcy Prediction13 references4 citations

TL;DR

This paper proposes a machine learning approach to predict credit risk in unsecured lending using a Light Gradient Boosting Machine (LGBM) classifier on imbalanced credit card datasets. After applying SMOTE to address class imbalance, LGBM outperformed six other models, achieving 95.53% accuracy and an AUC of 0.99, demonstrating superior performance in detecting credit defaults.

ABSTRACT

Since the 1990s, there have been significant advances in the technology space and the e-Commerce area, leading to an exponential increase in demand for cashless payment solutions. This has led to increased demand for credit cards, bringing along with it the possibility of higher credit defaults and hence higher delinquency rates, over a period of time. The purpose of this research paper is to build a contemporary credit scoring model to forecast credit defaults for unsecured lending (credit cards), by employing machine learning techniques. As much of the customer payments data available to lenders, for forecasting Credit defaults, is imbalanced (skewed), on account of a limited subset of default instances, this poses a challenge for predictive modelling. In this research, this challenge is addressed by deploying Synthetic Minority Oversampling Technique (SMOTE), a proven technique to iron out such imbalances, from a given dataset. On running the research dataset through seven different machine learning models, the results indicate that the Light Gradient Boosting Machine (LGBM) Classifier model outperforms the other six classification techniques. Thus, our research indicates that the LGBM classifier model is better equipped to deliver higher learning speeds, better efficiencies and manage larger data volumes. We expect that deployment of this model will enable better and timely prediction of credit defaults for decision-makers in commercial lending institutions and banks.

Motivation & Objective

• To develop a robust credit scoring model for unsecured lending using machine learning to improve default prediction.
• To address the challenge of data imbalance in credit card default datasets, where default instances are rare.
• To evaluate and compare the performance of seven machine learning classifiers in predicting credit risk.
• To identify the most accurate and efficient model for real-time deployment in commercial banking and lending institutions.
• To optimize model hyperparameters for improved generalization and scalability on large datasets.

Proposed method

• Applied Synthetic Minority Oversampling Technique (SMOTE) to balance the imbalanced credit card default dataset by generating synthetic minority class samples.
• Trained and evaluated seven machine learning classifiers: Logistic Regression, SVM, KNN, Decision Tree, Random Forest, XGBoost, and LGBM.
• Used Area Under the ROC Curve (AUC) and Accuracy Score as primary evaluation metrics to compare model performance.
• Employed hyperparameter tuning for LGBM using max_depth, learning_rate, and num_leaves to optimize model efficiency and reduce overfitting.
• Implemented a leaf-wise tree growth strategy in LGBM to improve training speed and accuracy compared to level-wise approaches.
• Conducted feature engineering on customer personal and credit history data to prepare inputs for model training and evaluation.

Experimental results

Research questions

• RQ1 Which machine learning model performs best in predicting credit defaults in unsecured lending with imbalanced data?
• RQ2 How effective is SMOTE in improving model performance on highly skewed credit default datasets?
• RQ3 What is the relative performance of LGBM compared to other gradient-boosting and traditional classifiers in credit risk prediction?
• RQ4 Can LGBM achieve high accuracy and efficiency on large-scale credit card datasets with limited memory usage?
• RQ5 What hyperparameter settings optimize LGBM performance for credit risk modeling in real-world banking applications?

Key findings

• The LGBM classifier achieved the highest accuracy score of 95.53%, significantly outperforming all other models tested.
• LGBM recorded an AUC of 0.99, indicating excellent discrimination ability between default and non-default cases.
• XGBoost ranked second with an accuracy of 91.96% and AUC close to LGBM, but LGBM showed superior training speed and memory efficiency.
• The SMOTE technique effectively mitigated data imbalance, enabling better learning from rare default cases.
• Among the evaluated models, LGBM demonstrated the best balance of speed, accuracy, and scalability for large-scale credit risk applications.
• Hyperparameter tuning of max_depth, learning_rate, and num_leaves further enhanced LGBM’s performance and reduced overfitting.