Machine Learning Credit Default Prediction Pipeline

Create a comprehensive Python machine learning pipeline for predicting credit default probabilities using scikit-learn and TensorFlow. The script must preprocess financial data from multiple sources, handle missing values through advanced imputation techniques, and develop a stacked ensemble model combining logistic regression, random forest, and gradient boosting classifiers. Implement cross-validation with stratified k-fold, generate detailed model performance metrics including ROC-AUC, precision-recall curves, and develop an interpretable feature importance visualization.