Interview Practice Questions

What does logistic regression architecture look like?

How is logistic regression trained?

Can logistic regression get stuck in local minimum?

How to understand if model is overfitted?

What regularization methods exist?

What metrics can be used for binary classification?

What is plotted on X and Y axes in ROC curve?

How will ROC-AUC change if all probabilities are divided in half?

Can we deploy model to production based on ROC-AUC 0.81 vs 0.80?

How to study model quality in more detail?

What model quality metrics do you know?

What baseline can you suggest building?

Why is baseline needed in ML model development?

What is MAPE and how is it calculated?

What methods can be used for training on imbalanced data?

What to do with highly correlated features?

Why do linear models perform poorly with imbalanced data?

What is the fundamental difference between SVM and logistic regression?

What do you know about ensemble methods?

How do CatBoost, LightGBM, XGBoost differ in tree construction?