Explainable AI Systems for Credit Scoring and Loan Risk Assessment in Digital Banking Platforms

This article reviews the use of explainable AI systems in enhancing credit scoring and loan risk assessment in online banking. We implemented four machine learning models—XGBoost, Random Forest, LSTM, and CatBoost—to determine how their accuracy can be compared for predicting creditworthiness. We utilized tools like SHAP and LIME in making the decision process lucid and transparent. The experiments show that XGBoost yielded the highest accuracy at $\mathbf{9 2. 5 \%}$, followed by LSTM at $\mathbf{8 9. 3 \%}$, CatBoost at $88.7 \%$, and Random Forest at $85.6 \%$. The hybrid model of LSTM and XGBoost achieved a dramatic boost in accuracy and interpretability and reached accuracy at $94.1 \%$. “In addition, the models proposed here displayed strong F1-score, recall, and precision values, with the hybrid model generating precision with a value of 0.91, recall with a value of 0.89, and $F 1$-score with a value of 0.90. The study also compared the proposed models with the current techniques and established breathtaking improvements over standard techniques. These findings confirm that Explainable AI models play a crucial role in establishing confidence and fairness in credit risk assessment systems by being action transparent and high performance”. The study also illustrates the capability of AI to support financial inclusion through offering responsible and accurate credit ratings to marginalized communities.