Enhancing Loan Approval Decisions through Machine Learning Algorithms

Résumé

This study presents academic state-of-the-art current research on using Machine Learning to
improve the predictability of bank loan approval procedures for financial inclusion. This study provides
solutions to several issues persisting in traditional methods of evaluating credit with standard credit assessment
procedures, including embedded bias, imprecision, and limited scalability across levels of socioeconomic status.
The methodology used in this article utilised a significant workflow to analyse the machine learning process
includes preparing the data, exploring it, creating new features, and system training and testing of nine
classification models, including both classical and ensemble learning methods. The outcomes specify that
ensemble learning models (Random Forest, Gradient Boosting, XGBoost and LightGBM) are capable of
providing very high accuracy with 98% accuracy, precision, recall, and F1 scores. By including additional data
sources and using explainable AI, ensemble models can demonstrate compliance with regulatory requirements
through their integrity, transparency, and consistency. The findings of this research show how ensemble models
can reduce classification errors, lower operational costs, and improve decision-making. Therefore, they have
the capability of decreasing operational costs and improving decision making may enable people in underserved
communities to have more access to financial services. The research highlights new ways that machine learning
can support greater financial inclusion and represents a scalable, trustworthy, and responsible approach for
leading banking institutions to broaden their ability to reach out to those who are not currently benefiting
from the availability of credit.