Dear Marie-Eve,
the last output of the LOO results is basically a weighted average between TPR and TNR.
The fact that we weigh TPR 3 times more than TNR relies on the assumption that having higher TPR is more important than higher TNR, i.e. we want to retain as much signal as possible while removing noise.
In general, when choosing the best threshold you need to reach a compromise between TPR and TNR depending on your needs/aims.
You can use the numbers derived from that formula if you retain TPR much more important than TNR, otherwise you decide the best trade-off by looking at TPR and TNR separately on in other combinations/weighted averages.
Hope it helps,
Ludovica
> On 27 Nov 2018, at 2:04 pm, Marie-Eve Hoeppli <[log in to unmask]> wrote:
>
> Hello,
>
> We ran the classifier training on our data with the LOO option to define which threshold we should use with our classifier.
>
> In the output of the LOO test, we can see the TPR and TNR, as well as the following formula: (3*TPR+TNR)/4
>
> We could not figure out what this formula describes and if we should also look at it when deciding the threshold or if we should only look at the TPR and TNR.
>
> Could you please let us know how to best proceed?
>
> Thank you,
>
> Marie-Eve Hoeppli
>
> ########################################################################
>
> To unsubscribe from the FSL list, click the following link:
> https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=FSL&A=1
########################################################################
To unsubscribe from the FSL list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=FSL&A=1
|