Improving Prediction of Credit Risks Based on Advanced Machine Learning and Feature Engineering Techniques in Banking Sector

The work exposes a resilient machine learning-based technique making use of the Gradient Boosted Tree (GBT) model for precise credit risk classification. The data from Lending Club, which is open to the public, is used. Preprocessing of this data consists of missing value imputation, redundant feature removal, and Min-Max normalization, in addition to label encoding. The Adaptive Synthetic (ADASYN) oversampling method is used to address the significant class imbalance. The Random Forest algorithm is selected for feature selection. The data is then split into training and testing sets. The Gradient Boosted Trees (GBT) model is trained with this data since it is very effective at handling structured financial data. Recall, accuracy, precision, ROC-AUC, and precision-recall curves are the primary evaluation tools for the model's performance. The GBT model, due to its characteristics, gives a very high level of 96% "accuracy" (acc), 95% "precision" (prec), 94% "recall" (rec), and 98% "F1-score" (f-measure) in contrast to the baseline models (Artificial Neural Network (ANN), K-Nearest Neighbors (KNN), and XGBoost). The discovered truth triumphs the entire process of the recommended technique for desiring credit risk assessment, thus allowing its introduction into financial decisions.