6. The confusion matrix for a machine learning model A is the following:

predicted

A B C

A 80 50 5

actual B 30 90 10

C 2 3 40

And the confusion matrix for a machine learning model B is:

predicted

A B C

A 80 60 35

actual B 10 80 10

C 2 3 30

Assuming that the cost matrix for the application domain is:

predicted

A B C

A 0 3 5

actual B 70 0 10

C 1000 90 0

Then

Which model, A or B, is better in terms of classification accuracy?

Which model, A or B, is better in terms of the total cost incurred for errors?