Exam DISC-DA Practice Question 212
Study guide: Exam DISC-DA
A homeowners wind model is built on five accident years of policy data, including a year with one large hurricane that produced thousands of claims. The analyst splits the records 70/30 at random into training and test sets, and the test-set lift is excellent. The chief actuary doubts the result. Which concern is most valid, and what should be done?
- A) The test set is too small; switch to a 90/10 random split to get more test claims
- B) Hurricane claims fall in both sets; test on a later period not used for training
- C) Random splits always overstate lift; abandon the test set and judge training fit
- D) The model has too few variables; add predictors until the training error stops falling
Answer this question and see the worked solution