1. Question 1

What does cross_val_predict(lr_model, X_train, y_train, cv=3) return?

• ❌ Predicted values of the test set using cross-validation.

• ❌ Computed a list of residual errors for the training set.

• ❌ Calculated the average R2 score from each fold.

• ✅ Predicted values of the training set using cross-validation.

Explanation:
cross_val_predict returns predictions for each training sample, generated by models trained on the other folds (not the final model).

---

2. Question 2

Correct way to define alpha list for ridge regression grid search?

• ✅ parameter = [{‘alpha’: [1,10,100]}]

• ❌ grid = alpha:[1,10,100]

• ❌ alpha = Ridge([1, 10, 100])

• ❌ parameter =[alpha: 1, 10, 100]

Explanation:
GridSearchCV expects a dictionary:
{'alpha': [values]} and wrapped inside a list.

---

3. Question 3

Model has 100-degree polynomial and R² = 0.99 on training. What next?

• ✅ Evaluate the model on the test datasets.

• ❌ Reduce the number of features before training.

• ❌ Use cross_val_predict on the training data.

• ❌ Select the model based on training score.

Explanation:
A high training R² with complex model suggests overfitting.
Check test performance to confirm.

---

4. Question 4

Why select ridge over standard linear regression?

• ✅ To reduce overfitting by penalizing large coefficients

• ❌ To remove irrelevant features from the model

• ❌ To increase the model’s flexibility and fit

• ❌ To reduce the model’s complexity and coefficients

Explanation:
Ridge regression adds L2 penalty, which shrinks large coefficients and helps control overfitting.

---

5. Question 5

Image description: Blue curve (model) is highly wiggly, matching each dot but not true line → This indicates:

• ❌ The model is a good fit

• ❌ No conclusion for the model

• ✅ It displays overfitting

• ❌ It displays underfitting

Explanation:
The model follows noise in the training data rather than the true function → classic overfitting.

---

🧾 Summary Table

---