Interview Practice Questions

Can ML model results be shown to customers?

How can external factors affect ML model?

What are alternatives to removing correlated features from the sample?

Can logistic regression overfit?

Why can CatBoost be more resistant to overfitting than logistic regression?

What to do if model overfits?

When should CatBoost be changed to logistic regression?

How did you use PD models? What algorithms did you apply?

What to do if model PSI exceeds 20-25%?

One model has Gini 48%, another 50%. Which to choose?

What is the role of model in credit application review process?

How do you organize the feature selection process to reduce the number of features from 300 to 40? Describe the steps and methods you use.

What hyperparameter optimization algorithms did you use?

What does the ML specialist's work process include in your team?

What are the criteria for choosing between logistic regression and XGBoost?

What statistical criterion underlies A/B testing of models?

How is the financial impact of customer churn prediction models calculated?

How is the target variable defined for PD (Probability of Default) models?

What models were used for credit scoring?

What methods for improving model quality were applied? Was stacking used?