Evaluating Classification Models

Question 1

What is the most used evaluation criterion?

Answer 1

Predictive accuracy (measurement of accuracy)

Question 2

What is K-fold cross-validation?

Answer 2

Divide data into K partitions, where you train on K - 1 number of partitions and test one 1 K testing set.

Question 3

How do you judge the accuracy of K-Fold CV?

Answer 3

Mean predictive accuracy over the K runs of the classification

Question 4

What is the advantage/disadvantage of K-fold CV?

Answer 4

Advantage: More reliable than single training/test set partition

Disadvantage: Computationally expensive