1
$\begingroup$

I am dealing with an unbalanced dataset. The total instances in my dataset is 1273 and the Yes class is 174 and No class is 1099. So the unbalance ratio is like 1:6. Now I know Recall measures how many yes classes we might have missed in the entire sample. I used Weka to classify the dataset. I was using Naive Bayes classifier and got recall 0.413. As per the definition of recall I can find out how many yes class I might have missed.

1273*41.3% = 525.749

However, I wonder how could I miss 525 yes classes where the number of yes classes itself 174.

Any help would be appreciated

$\endgroup$

1 Answer 1

2
$\begingroup$

I think this is just a confusion with the definition of recall. You can better remember it based on the positive class. When we talk about recall, we should look at the actual count of positive class in the dataset. In your case, if you take 'Yes' as the positive class, there are 174 entries in this class. In simple terms recall measures the ratio of actual positive class in the prediction against the actual positive class in the population whereas precision measures the ratio of actual positive class in the prediction against the total predicted positive class. Both case, only the denominator changes. Your model gives .413 as recall meaning out of the 1273 records in the population where 174 are the real positive class entries, 'x' number is classified as positive where x/174 = .413. This gives x = .413*174 = 71.862.

This means your model could classify only 71 as 'Yes' out of 174

I found this excellent video lesson for this - https://www.youtube.com/watch?v=2osIZ-dSPGE

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.