A Voting Ensemble Learning Model for Improved Credit Default Risk Prediction

Credit risk prediction is one of the most rudimentary pillars for practices in financial institutions. In order to maintain sound financial health and support the long-term safety of the institution, a strong framework for credit default risk evaluation is necessary. A strong credit default risk forecasting framework ensures financial institutions expand their lending activities and explore new markets. In different cities across the world, loan defaults in financial institutions result in adverse impacts such as financial losses, reduced profitability, and capital adequacy ratios. Therefore, it is high time to overcome traditional credit evaluation methods that are being followed in several financial institutions till date. This paper thoroughly evaluates the loan default prediction methodologies or similar assessment techniques proposed till date. Exploring all the best possible machine learning (ML) techniques, this paper proposes a soft-voting ensemble model combining three different (Random Forest, XG Boost, Gradient Boosting) and high-performing models to forecast credit default risks ensuring safety and success in institutional lending activities, evaluating creditworthiness of the borrower. The dataset was acquired from Kaggle. Following data preprocessing, the processed data was trained into the proposed ensemble model, followed by hyperparameter tuning, evaluation, and deployment of the model. The evaluation metrics include the F1 score, recall rate, accuracy, AUC-ROC and model precision. The final data visualizations portray the comparisons and the benefits of using this model across different financial institutions.