Does setting "prior" in the classification play a role in the classification algorithm or it just modifies observations in the prior classification step?

Based on the explanation in the "Algorithm" section here:

It seems that prior probablities are only used to remove datapoints that have zero prior probablity from training data and NOT playing role in the "classification process". Am I right? What does prior probablity is intended to do?

Answer 1

[Jalaj Gambhir]

No,

Prior Probabilities do affect the classification process. The step mentioned in the algorithm section only removes the training data corresponding to the classes for which the user has provided the prior probabilities as 0.

Prior probabilities are used to normalize the observation weights as stated in the next step here. I also recommend you to go through the references mentioned in the same link for detailed analysis on how SVM works.

Hope this helps!

[Zeynab Mousavikhamene]

@Jalaj Gambhir

Thanks for the responce. I assume if "prior probablity" affect the classification process, it should have mentioned somewhere in matlab when describing SVM like here. Well there are 6 refences some of them are books and I am not sure which part of these refences matlab used for "prior probablity" so it is very unclear where should I look for the answer to my question.

[Zeynab Mousavikhamene]

and would you please share your thoughts for this question:

https://www.mathworks.com/matlabcentral/answers/527169-is-svm-sensitive-to-unbalanced-observations-the-observations-in-one-class-is-3-4-times-of-the-obser?s_tid=prof_contriblnk