I explained that I would start with logistic regression for interpretability, then move to more complex models like Random Forest or XGBoost for performance. I also discussed feature engineering and evaluation metrics like ROC-AUC.