Comparative Performance Analysis of Machine Learning Algorithms for Non-Performing Loan Prediction

This paper investigates the application of machine learning models for Non-Performing Loan (N. P. L) prediction in the banking sector. The study evaluates the accuracy of diverse models, including traditional algorithms and advanced Machine Learning approaches such as Random Forest, Gradient Boost, XGBoost, LSTM, Decision Tree, Gaussian Naive Bayes and LightGBM. Training of the models are done on the dataset 'Training Data,’ available on Kaggle under the notebook named “Loan Prediction Based on Customer Behaviour”. Data processing and analysis of the given dataset was done on Kaggle's Jupyter Notebook cloud environment by using Python programming libraries. Through rigorous experimentation, it has been attested that Random Forest classifiers stands out to be best performing model when tested across all the performance benchmarks. This information thereby offers valuable guidance for institutions aiming to strengthen N. P. L prediction mechanisms and overall risk management strategies.