Re: [R] Reducing the FPR (false positive rate)

From: Liaw, Andy <andy_liaw_at_merck.com>
Date: Thu 16 Jun 2005 - 22:56:43 EST


> From: Anderson de Rezende Rocha
>
> Hello R-USERS,
>
> I think some people didn't understand my question. What I want is to
> use the training set to minimize the FALSE POSITIVE RATE.
>
> I think it is possible. I sacrifice ACCURACY to have less FALSE
> POSITIVES. I don't want a classifier result with 5% of FPR and, for
> example, 93% of ACCURACY. I want a 1% FPR sacrificing the 93%
> ACCURACY.
>
>
> Do you know how can I do this? I really need this because the
> requisite
> of the work I'm doing is only 1% of FPR.

I'm afraid it's you who's not getting the point. You don't need any classifier to get minimum FPR: just call everything `negative' and you have 0% FPR, which is as minimum as FPR gets.

FPR can not be fixed, as it's an unknown quantity. It make as much sense to say you want a classifier with a particular FPR as to say that you want a classifier with a particular error rate. What you can do is estimate the parameters of the classifier that gives you an _approximate_ FPR that's close to what you want, but there's no guarantee that you will get exactly that FPR on any test set.

Andy  

> Thanks and best regards.
>
> *************************************************
> |Üñð뮧µñ| - Anderson de Rezende Rocha
> Bacharel em Ciência da Computação - UFLA
> Mestrando em Ciência da Computação - UNICAMP
> Esteganografia e esteganálise digital
> UNIVERSIDADE ESTADUAL DE CAMPINAS, SP - BRASIL
> < http://andersonrocha.cjb.net >
>
> ______________________________________________
> R-help@stat.math.ethz.ch mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide!
> http://www.R-project.org/posting-guide.html
>
>
>



R-help@stat.math.ethz.ch mailing list
https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html Received on Thu Jun 16 23:07:46 2005

This archive was generated by hypermail 2.1.8 : Fri 03 Mar 2006 - 03:32:43 EST