In general, LOOCV requires fitting a model for a total of n times.

In general, LOOCV requires fitting a model for a total of n times.

• A. False
• B. True
• C. Only for certain models
• D. Never

Answer: (B)

Leave-One-Out Cross-Validation evaluates a model by holding out one observation at a time and training on all the remaining data. For each of the n observations, you fit the model to the other n−1 data points and then predict the held-out case. Repeat this for every observation, which means you perform a total of n fits. That is the essence of LOOCV. There are computational shortcuts for some models (for example, linear regression can use the hat matrix to compute LOOCV errors without refitting each time), but the standard procedure involves n separate fits. Hence the statement is true.

Leave-One-Out Cross-Validation evaluates a model by holding out one observation at a time and training on all the remaining data. For each of the n observations, you fit the model to the other n−1 data points and then predict the held-out case. Repeat this for every observation, which means you perform a total of n fits. That is the essence of LOOCV.

There are computational shortcuts for some models (for example, linear regression can use the hat matrix to compute LOOCV errors without refitting each time), but the standard procedure involves n separate fits. Hence the statement is true.