Predictive Credit Default Probability Machine Learning Pipeline

Construct a machine learning pipeline using scikit-learn and TensorFlow that predicts credit default probabilities with 95%+ accuracy. The model must integrate historical financial data, incorporate macroeconomic indicators, and use advanced ensemble techniques like gradient boosting and neural networks. Implement robust feature engineering, handle class imbalance, and create an interpretable model that financial regulators can audit. Include cross-validation strategies and a comprehensive model performance evaluation framework.