Maxent accuracy assessment

[Hira Fatima]

Hello again:

i removed some variables on the basis of correlation analysis, but the accuracy of my results is decreasing, it was higher for the one with all variables, i used 2 study areas, for one it was giving similar results (like didn't improve here either), for other it was decreasing.. it that possible or am i going in wrong direction...

[Hira Fatima]

Also if  i want to create response curves on my own, does maxent hold any .csv file for that??

[Husam El Alqamy]

Select the option of "write plots data" in maxent and it will produce dat text files of the response curves which you can bring into excel or any other software or even into R.

Regards

On Thu, Jul 3, 2014 at 10:20 AM, Hira Fatima <fatim...@gmail.com> wrote:

Also if  i want to create response curves on my own, does maxent hold any .csv file for that??

--
You received this message because you are subscribed to the Google Groups "Maxent" group.
To unsubscribe from this group and stop receiving emails from it, send an email to maxent+un...@googlegroups.com.
To post to this group, send email to max...@googlegroups.com.
Visit this group at http://groups.google.com/group/maxent.
For more options, visit https://groups.google.com/d/optout.

--

Husam El Alqamy, B.Sc., M.Phil.
Sr. Biodiversity GIS Analyst ,
Environmental Information Sector, EIS
Environmental Agency Abu Dhabi,UAE
Antelope Specialist Group, ASG - IUCN

[Martin Damus]

Hi,

What do you mean by 'the accuracy of my results'? Do you mean that the predicted distribution is now less narrow, or it is extending into parts you know are not suitable based on other information, or is the AUC decreasing? All of the above will happen when you reduce the number of variables because the model has less information on which to base its prediction. But more of the same information (correlation) doesn't necessarily improve the prediction, it may narrow the final output, but only because of the bias introduced by the duplicated information in the correlation. That said, Maxent is pretty good at dealing with correlations, so don't remove only somewhat-correlated variables. I see publications where only variables correlated at 75% or greater are removed. It might be better to find other predictors, such as forest type, agricultural crop, population density, other factors that can be mapped that you believe influence the distribution of your subject organism, rather than including only the generic climate data.

Martin

---

From: Hira Fatima <fatim...@gmail.com>
To: max...@googlegroups.com
Sent: Thursday, July 3, 2014 1:59:51 AM
Subject: Maxent accuracy assessment

Hello again:

i removed some variables on the basis of correlation analysis, but the accuracy of my results is decreasing, it was higher for the one with all variables, i used 2 study areas, for one it was giving similar results (like didn't improve here either), for other it was decreasing.. it that possible or am i going in wrong direction...

[Hira Fatima]

the predicted distribution got a little wider with few uncorrelated variables, but thats ok, with all variables distribution was a little narrow, that according to cruz-cardenaz et al 2013, is because with increasing variables maxent has more rules to follow so range becomes narrow.... here i was referring to AUC i know AUc is not a good measure of accuracy but i also tried running TSS, which was mentioned sometime ago on this forum, but with both accuracy decreased with reducing the variables, maybe while removing collinearity some ecologicaly important variable got removed, but then again we go all the way to remove collinearity to improve accuracy, then whats the point if the very thing we are trying to retain decreases... and moreover important variables might also get deleted.... please explain

--
You received this message because you are subscribed to a topic in the Google Groups "Maxent" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/maxent/oR5UlrVcvVs/unsubscribe.
To unsubscribe from this group and all its topics, send an email to maxent+un...@googlegroups.com.

[Hira Fatima]

no sorry i think i misinterpreted According to this study 2013, "The models of distribution generated with raw environmental predictors generally predict a smaller occupancy surface. That is so, because these models must satisfy a large number of rules, so that, a given record can be classified as a true presence"' but if thats the case with low predictive accuracy model is showing higher covered areas, the areas where species might not be prevalent... in both cases i argue that these methods of removing collinearity doesnt yield better accuracies then those that ignore collinearity rather sometimes decreased as in mine...