Credit Risk Prediction in Digital Finance Ecosystems Using Ensemble Learning, Deep Neural Networks, and Sentiment Fusion

This study puts forth a hybrid AI-driven framework for predicting credit risk in digital finance ecosystems by amalgamating structured financial data with unstructured market sentiment. A feature fusion strategy was used with the Kaggle Home Credit Default Risk dataset (307,511 records, 122 features) and sentiment scores from news and social media that were generated using NLP models. SMOTEENN was used to fix the class imbalance, and regularized logistic regression was used to choose the best features. We made and tested an ensemble machine learning stacking model (XGBoost, LightGBM, Random Forest) and a CNN-LSTM deep learning architecture using ROC-AUC, F1-score, precision, and recall. The experimental results show that the stacking ensemble did much better than the baseline structured-only models (accuracy: 95%, ROC-AUC: 0.9817). The results show that sentiment fusion greatly improves the accuracy and reliability of systems for assessing credit risk.