Stacking Ensemble Approach for Predicting Loan Approval Using Machine Learning Techniques

Digital lending and fintech innovations have upended established banking systems, changing financial inclusion and credit availability in nations around the world. This study examines how peer-to-peer (P2P) and digital lending platforms are changing, emphasizing how technologies like artificial intelligence and machine learning are changing the way loans are approved. A thorough study of the literature highlights the opportunities and problems in the digital lending ecosystem, such as algorithmic risk assessment, customer trust, financial exclusion, and regulatory loopholes. This paper suggests a strong machine learning approach that uses a stacking ensemble model to accurately forecast loan approvals in order to address these issues. The data was pre-processed using train-test partitioning, exploratory analysis, and label encoding using a publicly accessible Kaggle dataset that included applicant demographics, financial characteristics, and credit histories. With XGBoost serving as the meta-learner, the ensemble incorporates the Gradient Boosting Model, Efficient Gradient Boosting, AdaBoost, and Extra Trees classifiers as base learners. With an accuracy of 98%, the model was assessed using measures including accuracy, precision, recall, F1-score, and error metrics (MAE- Mean Absolute Error, MSE- Mean Squared Error, and RMSE- Root Mean Square Error). According to correlation studies, factors including assets, income, and CIBIL scores have a significant impact on loan approvals. Outperforming conventional methods, the model showed balance and generalization across both classes. The usefulness of these models for automated, data-driven credit determinations is emphasized in the paper's conclusion.