Exam SRM Practice Question 2
Study guide: Exam SRM
An actuary randomly splits 500 claim records into a training set of 250 and a held-out test set of 250. She fits polynomial regressions of claim severity on driver age, of degree 1, 2, 3, 5 and 8, to the training set, and computes each model's mean squared error (MSE) on both sets:
| Degree | Training MSE | Test MSE |
|---|---|---|
| 1 | 41.2 | 43.0 |
| 2 | 30.5 | 32.8 |
| 3 | 24.9 | 27.1 |
| 5 | 23.6 | 28.4 |
| 8 | 21.8 | 33.9 |
Which conclusion about the choice of degree is best supported?
- A) Degree 8 should be chosen, because it has the lowest training MSE of the five models
- B) Degree 1 should be chosen, because its test MSE is closest to its training MSE
- C) Degree 3 should be rejected, because a test MSE above the training MSE shows it overfits
- D) No degree can be chosen yet, because the training MSE has not stopped falling
- E) Degree 3 should be chosen, because above it the test MSE rises as the fit chases noise
Answer this question and see the worked solution