Loan-Default-Prediction

• This project aims to provide a robust and accurate solution to the loan defaulting problem, helping financial institutions make more informed lending decisions and reduce their risk exposure.
• Appropiate EDA, data balancing by SMOTE and fitting Logistic, SVM, Random Forest, LightGBM, XGboost, ANN with hyperparameter tuning with best AUC 0.938 in Logistic Reg. and ANN and 89.43% accuracy in Light GBM.
• From Contingency table of employment and defaulted Pearson’s χ 2 test for independence gives p-value approximately 0.0005 << 0.05 indicating high dependence. Bank balance and employment should be checked before lending.