Loan Default Prediction System using Machine Learning for Financial Institute In Sri Lanka

Abstract

Lending is an active investment for financial institutions. One of the most important things that financial institutions need to remember these days is credit risk management. Even though lending money is profitable for lenders, financial institutions risk suffering significant capital losses in the event that a loan default. For financial institutions, it will be crucial to forecast loan and lease defaulters in order to estimate default risk in advance. Therefore, rather than relying on conventional credit scoring techniques, it is crucial to deploy precise and scientific data-driven solutions to mitigate this financial risk. Lenders will be able to estimate the creditworthiness of borrowers using this suggested method before approving a loan or lease for the clients. There are currently no AI-powered machine learning solutions available in Sri Lanka, despite the fact that several financial institutions are already adopting them. To control financial risks for Sri Lanka's leasing firms and financial institutions, an end-to-end solution for loan default prediction is presented. For institutions to be financially stable, prompt loan default prediction is essential. The impartiality and dependability of current methodologies are frequently lacking, which could cause intervention delays. This study offers a machine learning-based early loan default detection tool designed especially for financial institutions. With careful data preprocessing and the use of multiple techniques, including Random Forests, Decision Trees, and Support Vector Machines, the model seeks to improve robustness and accuracy. After undergoing a thorough review, the suggested loan default prediction model shows good accuracy and interpretability, making it a potentially useful tool for financial institutions to manage and reduce the risk of loan default. Financial institutions must foresee loan defaults in order to reduce risk and maximize lending choices. The creation and comparison of machine learning models for forecasting the probability of loan default is the main goal of this research project. In order to predict loan default, the study assesses the effectiveness of five distinct machine learning algorithms: RandomForestClassifier, LogisticRegression, SVM, KNeighborsClassifier, and DecisionTreeClassifier. The models are trained and tested using actual data from financial institutions. The project's goal is to determine the best algorithm for precise loan default prediction through extensive testing and research. The research's conclusions improve risk management procedures in the banking industry and give decision-makers useful information for reducing the likelihood of loan default.