Aug 2026· International Scientific Journal of Engineering and Management· Vol 05, pp. 1-9· 0 citations
TL;DR
Overall, it can be concluded that the application of gradient-boosted tree ensembles in combination with good feature engineering and hyperparameter optimization makes it possible for such models to provide an optimal trade-off between accuracy, speed, and explainability when predicting housing loan default.
Abstract
Abstract - The timely and precise assessment of housing (mortgage) loan default risk is crucial for lenders to be able to minimize their credit losses while, at the same time, maintain fairness and efficiency in loan approval. Conventional underwriting methods are based on the calculation of various ratios and judgment of an underwriter. However, this is an extremely difficult process that can take a long time and does not provide consistency between different underwriters. The goal of this research is to analyze how machine learning can be used as the full process of predicting a housing loan default by means of using client’s demographic information, income, credit and property data. For this study, five classification algorithms were trained: Logistic Regression, Decision Tree, Random Forest, Gradient Boosting, and XGBoost. Every numerical feature of the dataset was median-imputed and standardized while every categorical feature was mode-imputed and one-hot encoded. Every model was tuned using RandomizedSearchCV with stratified cross-validation with 5 folds based on F1 score. And the last engineered features, Debt-to-Income Ratio and Loan-to-Value Ratio, were obtained via the use of applicant’s income, existing debts, loan amount, and property value.The XGBoost model produced the best performance in that it got 95.08% of accuracy, 96.62% of precision, 97.54% of recall, and 0.9708 of F1 score while training in 78.8 seconds, which is a record speed compared to Gradient Boosting which took 984.9 seconds and Random Forest requiring 458.0 seconds. The model was deployed through a Flask application with features such as user sign-up, session-based authentication, an explanation engine indicating the main influences of particular predictions but also via a command-line application that runs offline predictions. Overall, it can be concluded that the application of gradient-boosted tree ensembles in combination with good feature engineering and hyperparameter optimization makes it possible for such models to provide an optimal trade-off between accuracy, speed, and explainability when predicting housing loan default.
Keywords: Housing Loan Default Prediction, Credit Risk Modelling, Machine Learning, XGBoost, Random Forest, Gradient Boosting, Debt-to-Income Ratio, Loan-to-Value Ratio, Flask Web Application.
Credit risk classification is an important component of financial decision-making because inaccurate classification may increase payment-default exposure and reduce lending quality. Credit datasets commonly contain numerical variables with different measurement scales and imbalanced distributions between default and non-default clients, which can affect the reliability of machine-learning models. Objective: This study aims to develop and evaluate a Support Vector Machine model for classifying credit card clients into default and non-default categories. The study also examines the influence of numerical standardization, kernel selection, hyperparameter optimization, and balanced class weighting on classification performance. Methodology: A quantitative experimental approach was applied using the Default of Credit Card Clients dataset from the UCI Machine Learning Repository. The dataset consisted of 30,000 observations and 23 predictor variables. Data were divided into training and testing subsets using a stratified 80:20 ratio. Categorical variables were encoded, numerical variables were standardized, and several SVM kernels were evaluated. Hyperparameter selection was conducted using five-fold cross-validation. Model performance was assessed using accuracy, precision, recall, F1-score, specificity, balanced accuracy, and ROC–AUC. Findings: The SVM model trained without standardization failed to identify default clients effectively. Numerical standardization substantially improved classification performance, while the radial basis function kernel produced the strongest validation results. The selected balanced RBF-SVM achieved 77.13% accuracy, 48.54% precision, 56.22% recall, 52.09% F1-score, 83.07% specificity, 69.65% balanced accuracy, and 75.10% ROC–AUC. Balanced class weighting improved default detection but increased false-positive predictions. Implications: The model can support financial institutions as an initial credit-risk screening tool. Its predictions should be combined with document verification, repayment-capacity analysis, and manual assessment rather than being used as the sole basis for credit approval. Originality: This study provides a controlled evaluation of SVM performance by integrating feature standardization, kernel selection, hyperparameter optimization, class-imbalance treatment, and class-sensitive performance metrics. The study demonstrates that credit-risk models should be selected based on balanced default detection rather than overall accuracy alone.
Andre Pratama Adiwijaya· International Journal for Sc...· 0 citations
Non-Performing Loans (NPL) are a fundamental indicator of a financial institution's asset health, reflecting loans that fail to meet interest or principal payment obligations as agreed. A high NPL ratio negatively impacts a bank's financial performance, such as decreased profitability as measured by Return on Assets (ROA) and decreased liquidity. Bank Indonesia sets an NPL tolerance limit of 5% of total credit provided by banking financial institutions. Therefore, a predictive model is needed that can detect the possibility of customers experiencing NPLs early. This study aims to identify relevant factors in predicting NPLs and create an NPL prediction model based on these factors. The contribution of this study lies in combining the results of three feature selection techniques: Chi-Square, Mutual Information, and Random Forest feature importance, using the average score eliminated by the Recursive Feature Elimination technique. Several ensemble algorithms, namely Random Forest, XGBoost, Gradient Boosting, and LightGBM, were explored to produce the best-performing model. Then, hyperparameter tuning was performed on the best model. The Random Forest model produced the best performance, with 92.17% accuracy, 78.1% precision, 98.1% recall, and 95.5% AUC. Hyperparameter tuning was shown to improve recall, thus improving the model's ability to measure how much positive data (Current class) was successfully predicted by the model. The results of this study can assist management in making credit decisions. Thus, it is hoped that it can help reduce the number of NPL cases.
David Jefri Aruan, Rusdah Rusdah, Ahmad Pudoli· IDEALIS : InDonEsiA journaL...· 0 citations
This study explores the use of Explainable Artificial intelligence techniques to improve the interpretability of credit default prediction and highlights the practical value of explainable machine learning in developing more understandable, trustworthy, and accountable credit risk assessment systems for real-world financial decision-making.
Muskan, B. Sidhu· International Journal of Com...· 0 citations
The study focused on developing a creditworthiness prediction model utilizing artificial neural network. Credit risk evaluation has a relevant role for financial institutions, as lending could result in real and immediate losses. In particular, default prediction was one of the most challenging activities in managing credit risk. The objective was to enhance the accuracy and reliability of credit risk assessments by leveraging the computational power and learning capabilities of artificial neural networks. The parameters of the dataset include the customer's place of work, loan history, monthly salary, loan amount, transaction history, and credit history, all stored in the trained database. When a customer comes to apply for a loan, the system checks if the user is qualified based on these parameters and then approve or disapprove the loan accordingly. Rigorous testing and validation were conducted to ensure the model's robustness and generalizability. The results demonstrated that the neural network-based model significantly outperformed traditional statistical methods, providing more precise predictions of creditworthiness. An Object-Oriented Analysis and Design Methodology (OOADM) approach, which incorporated Unified Modeling Language for analysis and design, was used. The development stage was completed using a set of software tools, including Python and the MySQL database system.
E. C., Uzo Blessing Chimezie, Ukekwe Emmanuel C· International Journal of Lat...· 0 citations
This study addresses the prediction of credit default risk using the Home Credit Default Risk dataset. Due to the approximately eight percent imbalanced class distribution in the data structure, an intensive feature engineering pipeline that reduces multiple tables to the customer level was applied before modeling. The implementation was designed to include behavioral summaries generated from external credit history, previous applications, installment behavior, credit card transactions, and POS cash records. Missing value handling, encoding, and multicollinearity reduction steps were performed on the obtained features. LightGBM was selected as the classification model, and the hyperparameters were optimized with Optuna under cross validation. In the best configuration, the mean AUC value was observed as 0.78682 and the mean PR AUC value as 0.28015. In order for the probabilities to be used more reliably in decision making processes, Platt scaling and SHAP based explainability analysis were applied.
Büşra Kurun, Okan Bursa· Signal Processing and Commun...· 0 citations
Large volumes of loan applications motivate automated decision-support systems that can reduce processing delays and improve consistency while controlling credit risk. This study presents a unified supervised-learning framework comparing XGBoost, Gradient Boosting, and CatBoost for loan approval prediction. The experiments use the Dream Housing Finance dataset containing 614 applications and 12 predictive variables after removing Loan_ID. The pipeline includes missing-value treatment, feature engineering, scaling, SMOTE-based class balancing applied only to training data, and evaluation on a held-out test set of 169 samples. Perfect training performance is treated as a diagnostic warning rather than evidence of generalization. CatBoost achieved the best held-out accuracy (88.17%), precision (88.37%), recall (88.37%), and F1-score (88.37%), with 10 false approvals and 10 false rejections. Confusion-matrix analysis, false-positive and false-negative rates, balanced accuracy, and Wilson confidence intervals indicate the most balanced performance among the evaluated models. The framework is intended as a prototype decision-support approach; larger multi-institutional validation, probability-based discrimination analysis, explainability, calibration, and fairness assessment are required before deployment in real lending environments.
N. Dandotiya, Kirti Jain, Prashant Kumar Shrivastava· 2026 International Conferenc...· 0 citations
