0
$\begingroup$

These past days, in college, we have been learning about NaiveBayes. Since it's a classification algorithm, I was wondering if I could evaluate NaiveBayes models the same way (using the same metrics) we evaluate other classification algorithms like SVM, LogisticRegression or DecisionTrees.
It makes sense to me to use metrics like precision, recall or F1-score to evaluate it. But I am doubting with metrics like the ROC curve or the PR-curve? Would it be correcto to evaluate my model using these type of curves? Or it doesn't make sense to build a ROC/PR-curve for NaiveBayes models?

Thanks a lot! :)

$\endgroup$

2 Answers 2

1
$\begingroup$

Using metrics makes sense in the context of the task, not algorithms. After all, we do not train algorithms on quality metrics, but compare different models by quality metrics.

That is, if this is a task, for example, ranking in credit scoring (giving more money to the best, and not giving it to the worst), then the area under the roc auc curve shows the quality of ranking and makes sense. If the task is to predict a disease during an epidemic , then recall is a good choice . Etc. I'm sure you can use the roc auc metric for any models that are capable of making continuous predictions, including for Naive Bayes classification.

$\endgroup$
0
$\begingroup$

As @Andrew mentions, the evaluation metrics you're going to be using are task-specific, and not algortihm-specific. Your task is classification, and the algorithm you're using is Naive-Bayes. Precision, recall, F1, AUC, etc. are all classification evaluation metrics, and it is sensible that you can use any (and even all together) to evaluate your model.

$\endgroup$

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.