Question about the unexpected no data in training gain, test gain, and AUC

[Xinyuan Cao]

The environment variable of "Topsoil_Gravel_Content" showed no data in training gain, test gain, and AUC in the average result of 10 replications (Fig. 1). However, the “Topsoil_Gravel_Content" can be successfully read by Maxent model, and there is value exists in each replicated model (Fig.2).

I'm not sure whether this phenomenon would affect the prediction results of species distribution suitability. Any recommendation would be sincerely appreciated!

Fig.1 Average result of 10 replications in training gain

Fig.2 Average result of the first replicated model in training gain

[Jamie M. Kass]

This test shows variable importance, not predictions from the final model. One variable is removed, the model is fit, then predictions are made. These predictions are compared with a model fit using only the withheld variable. The result is a measure of variable importance and would not reflect predictions made by your final model using all the variables. Not sure why the avg of the 10 reps shows no gain though -- might be a bug?

Jamie