Average ROC curves for binary classification using a threshold

Hi All,

I am new to machine learning. My goal is to generate a threshold from the data itself which can guide for the classification of data into 2 groups (i.e group a =below thresh, group b=above thresh). I am plannig to use ROC analysis to describe performance of such threshhold. However, as I was reading the ROC literature, I found that probabilities generated for a model is used as threshold to classify the data and generate ROC curve and not the threshhold from data itself. But, I would like to use the thresholds generated from the data itself and perform the ROC ananlysis. On top of this, I would like perform a 10fold cross validation on this ROC analysis. Can anyone suggest the procedure or guide me to correct reference?

Regards

Harshan Ravi

[Aditya Patil]

Can you clarify on what you mean by thresholds generated from data? Do you want to get a single threshold value that classifies the data? If so, why do you need ROC analysis(as that is intended for analysing different threshold values)?

[KALYAN ACHARJYA]

@Harshan Ravi What type of data you have?