Predictive Modeling Flashcards

Which regularization technique sets some regression coefficients exactly to zero, effectively performing feature selection?

Answer: LASSO (L1)

LASSO regression uses an L1 penalty that drives some coefficients to exactly zero, automatically selecting the most relevant features.

When building a propensity-to-buy model, the target variable is typically:

Answer: Binary — purchased or did not purchase

Propensity-to-buy models predict the probability of a binary outcome: whether a customer will purchase (1) or not (0).

What does an AUC-ROC score of 0.5 indicate about a classification model?

Answer: The model performs no better than random chance

An AUC of 0.5 means the model's ROC curve follows the diagonal, indicating it performs no better than random guessing.

In gradient boosting, how are subsequent models trained?

Answer: On the residual errors of previous models

Gradient boosting trains each new model to correct the residual errors (pseudo-residuals) left by the combined ensemble of previous models.

A marketer wants to predict next month's email campaign revenue. Which type of model is most appropriate?

Answer: Regression

Predicting a continuous value like revenue requires a regression model, not a classification model.

What is the key advantage of using cross-validation over a single train/test split?

Answer: It provides a more reliable performance estimate by using all data for both training and testing

Cross-validation rotates the test set across multiple folds so every observation is used for testing, yielding a more reliable and less variance-prone performance estimate.

Which of the following best describes 'feature engineering' in predictive modeling?

Answer: Creating or transforming input variables to improve model performance

Feature engineering involves creating new variables, transforming existing ones, or combining features to give the model better predictive signals.