An Enhanced Ensemble-Based Framework for Loan Approval Prediction Using Machine Learning

Abstract

Loan approval prediction is an important application in the banking and finance industry to automate and make easy the loan application decision process. In this work, we introduce the use of machine learning algorithms, namely Random Forest and

Decision Tree classifiers, to predict if a loan application should be approved or not based on applicant information. Features used in the dataset are income, education, employment, CIBIL score, loan amount, and dependents. Preprocessing methods like data cleaning, feature selection, and label encoding for categorical variables were used. Train-test split method with stratification was used for training and testing the models in order to keep the classes balanced. Model performances were checked based on accuracy, classification report, and confusion matrix. Our findings indicate that Random Forest classifier has a better performance than the Decision Tree model as far as accuracy and generalizability are concerned. The Random Forest model was trained on processed features and pickled using the pickle library to deploy in a real-world system for loan approvals. These results prove that techniques such as Random Forest can efficiently improve the precision and speed of loan approval mechanisms while reducing the level of human bias.