SVM Classifier Evaluation in R

Question

I am working on an application to evaluate the performance of SVM on a number of different datasets.

SVM is working well, yielding 98% diagonal accuracy.

I now need to evaluate the performance of this model, so using ROCR library, I plotted the ROC, err & acc curves.

The ROC is displayed as the False Positive Rate vs. True Positive Rate, and for the error & accuracy curves, these are plotted against an element called cutoff.

Could someone explain what this cutoff is exactly? (Needless to say, I am relatively new to this subject)

Thanks in advance

Answer 1

Usually, the predicted values in classifiers for a binary response variable is interpreted as a probability which shows the observation belongs to class 1 or not.

To make a decision at this point, a cutoff value is used, i.e. if the probability is greater than the cutoff, it belongs to class 1 and if the probability is lower than cutoff it, belongs to class 2.

For more information, please see this free book: "The Elements of Statistical Learning".