Interpretable Machine Learning Techniques for Causal Inference Using Balancing Scores as Meta-features.