IBM interview question

How do you handle data leakage? How do you select features for a model? Difference between L1 and L2 regularization. How do you handle multicollinearity? When would you choose precision over recall? Explain overfitting and how to prevent it. How do you tune hyperparameters? Cross-validation vs train-test split. How do you handle outliers? How does XGBoost work internally? How do you version models and data?