Scopus Indexed Publications

Paper Details


Title
A Hybrid XGBoost-MLP Approach for Credit Scoring and Risk Prediction in Microfinance

Author
, Md. Jobayer Ahmed, S.M. Tanvir Anjum Fahim,

Email

Abstract

The paper will investigate how machine learning models can be useful in credit scoring and risk analysis at microfinance institutions. This research will examine the effectiveness of various machine learning algorithms (XGBoost, Random Forest and Multi-Layer Perceptron-MLP) in loan default prediction and identification of the most powerful variables in credit risk. The modeling program performance on a dataset of the UCI Machine Learning Repository is evaluated in terms of accuracy, precision, recall, F1-score, and AUC-ROC as the performance measures. XGBoost has a better result with a precision of 94.50, accuracy of 93.23, recall of 73.77, F1-score of 82.86, and an AUC-ROC value of 94.90. Random Forest model has an accuracy of 92.83 and AUC-ROC of 93.49 as opposed to the lower accuracy of 91.87 in MLP model. The XGBoost-MLP hybrid model shows better performance with an AUC ROC of 95.00. The model is explained with the use of the SHAP and the LIME methodologies, thus making informed decisions in financial settings. The results indicate that relevant credit-risk analysis would allow machine learning systems to enhance efficiency in credit risk reduction and enhance financial inclusion. Additional research questions should also enable the analysis of the ethical concerns regarding the privacy and fairness of data and explore the impact of more unstructured data on improving model accuracy.


Keywords

Journal or Conference Name
2025 28th International Conference on Computer and Information Technology, ICCIT 2025

Publication Year
2025

Indexing
scopus