Interpretable deep learning-based analysis for liquidity risk forecasting in American banks

This study examines the use of interpretable deep learning and traditional machine learning models for forecasting liquidity risk in American banks. It compares the performance of models such as Long Short-Term Memory (LSTM) networks, Gated Recurrent Units (GRUs), Convolutional Neural Networks (CNNs) with traditional approaches like Random Forest, Logistic Regression, K-Nearest Neighbors (KNN), and Gradient Boosting in terms of accuracy, precision, and ROC AUC. The study emphasizes the application of the Local Interpretable Model-agnostic Explanations (LIME) technique to enhance the interpretability of predictions. The results indicate that while deep learning models provide superior accuracy and precision, LIME's integration offers valuable insights into the decision- making processes, crucial for regulatory compliance and informed risk management Keywords: Interpretable Deep Learning, traditional Machine Learning.