Complex Credit Risk Predictive Model Architecture

Design a comprehensive machine learning pipeline for predicting loan default probabilities using heterogeneous data sources. Create a modular architecture that integrates historical loan performance data, macroeconomic indicators, and alternative credit scoring signals. Develop a feature engineering strategy that handles missing data, creates interaction terms, and implements robust cross-validation techniques specifically for financial risk assessment. Include detailed model interpretability mechanisms and explain how different feature categories contribute to default prediction accuracy.