Hello everyone,

I am writing to ask you about a concept that is not clear to me. I am doing a study of a disease where I have a cohort of patients vs healthy. In this study I use 4 genes as predictors. The models I am analyzing are logistic regression and SVM.

My objective is to evaluate which of these 4 genes works best as a predictor in each model. So I draw the parameters Accuracy, Precision, Recall, Specificity, F1 Score, Confusion Matrix, ROC curves and AUC.

What I have learned and what I read about ROC curves to measure the diagnostic prediction of a gene, is that the closer to 0.8-1 it is, it could be considered a possible good predictor. However, I am doing a comparison between 4 genes and some colleagues have recommended that in addition to the AUC value, I calculate the significance of the ROC curve to see which of the 4 curves is more significant in each model. That is, to see which of those AUC is more significant.

I am trying to find more information about it but I am confused, is it correct to do that, does it apply in machine learning, how would it be done?

Thank you very much for your help,