Hybrid Loan Default Prediction System Using Machine Learning and Deep Neural Network (DNN)

Accurate loan default prediction is crucial for managing credit risk. Traditional models struggle with class imbalance, where defaults are rare. This study creates a hybrid system combining machine learning with Deep Neural Networks, using Adaptive Synthetic Sampling (ADASYN) to effectively address this imbalance. The research utilized a comprehensive dataset of 255,347 loan records with 18 predictive features, including borrower demographics, financial attributes, and loan characteristics. The study employed rigorous exploratory data analysis, four models were systematically evaluated: Logistic Regression, Random Forest, XGBoost, and the proposed Hybrid ML-DNN system with ADASYN integration. The study's methodology involved preprocessing data, applying ADASYN to handle class imbalance, and creating a hybrid model. This combines interpretable traditional ML with powerful deep neural networks to better predict complex, non-linear patterns in loan defaults. Traditional baseline models (Logistic Regression, Random Forest, XGBoost) achieved misleadingly high overall accuracies (88%+) but failed catastrophically at default detection, with recall rates between 0% and 8% for the minority class. These models essentially learned to predict all loans as "good," rendering them operationally ineffective for risk assessment. In stark contrast, the Hybrid ML-DNN model with ADASYN achieved a 52% recall rate for defaults a more than six-fold improvement over the best baseline model while maintaining a 79.83% overall accuracy and 0.7571 ROC AUC score. The model successfully identified 3,074 actual defaults out of 5,931 total defaults in the test set, compared to baseline models that identified fewer than 500 defaults. However, this improvement came with a precision trade-off of 29% for the default class, resulting in 7,442 false positives, highlighting the inherent business decision between risk mitigation and customer acquisition. The study proves ADASYN is essential for effective loan default prediction, not just an enhancement. It provides a robust, transparent AI framework for financial institutions, enabling improved risk assessment and promoting more stable and responsible lending practices

Loan default prediction; Machine learning; Deep Neural Network (DNN); Hybrid model; ADASYN; Class imbalance; Credit risk assessment

1. Banasik, J., Crook, J. and Thomas, L. (2003). Sample selection bias in credit scoring models. Journal of the Operational Research Society, 54(8), pp.822–832. doi: https://doi.org/10.1057/palgrave.jors.2601578. [Google Scholar] [Crossref]

2. Buda, M., Maki, A. and Mazurowski, M.A. (2018). A systematic study of the class imbalance problem in convolutional neural networks. Neural Networks, 106, pp.249–259. doi: https://doi.org/10.1016/j.neunet.2018.07.011. [Google Scholar] [Crossref]

3. Duan, J. (2019). Financial system modeling using deep neural networks (DNNs) for effective risk assessment and prediction. Journal of the Franklin Institute, 356(8), pp. 4716–4731. https://doi.org/10.1016/j.jfranklin.2019.01.046. [Google Scholar] [Crossref]

4. Haibo He, Yang Bai, Garcia, E.A. and Shutao Li (2008). ADASYN: Adaptive synthetic sampling approach for imbalanced learning. [online] IEEE Xplore. doi:https://doi.org/10.1109/IJCNN.2008.4633969. [Google Scholar] [Crossref]

5. Hossain, M.R. et al. (2025). Enhancing Credit Scoring with Multimodal Deep Learning: A Hybrid Neural Network Approach Using Structured and Unstructured Financial Data, International Interdisciplinary Business Economics Advancement Journal, 06(07), pp. 09–22. https://doi.org/10.55640/business/volume06issue07-02. [Google Scholar] [Crossref]

6. Khandani, A.E., Kim, A.J. and Lo, A.W. (2010). Consumer credit-risk models via machine-learning algorithms. Journal of Banking & Finance, 34(11), pp.2767–2787. doi:https://doi.org/10.1016/j.jbankfin.2010.06.001 [Google Scholar] [Crossref]

• What the Desert Fathers Teach Data Scientists: Ancient Ascetic Principles for Ethical Machine-Learning Practice
• Comparative Analysis of Some Machine Learning Algorithms for the Classification of Ransomware
• Comparative Performance Analysis of Some Priority Queue Variants in Dijkstra’s Algorithm
• Transfer Learning in Detecting E-Assessment Malpractice from a Proctored Video Recordings.
• Dual-Modal Detection of Parkinson’s Disease: A Clinical Framework and Deep Learning Approach Using NeuroParkNet