I have used logistic regression for prediction of Churn and the sensitivity vs 1-specificity curve is coming out to be:
From here it can be seen that the cutoff value is around 10%.Is there a way to find out the exact value in the graph,something like abline??
Also,the ROC curve obtained via:
ROC(form = Churn ~ .,plot = c(“sp”,“ROC”),
PV = T,MX = T,MI = T,AUC = T,data = logit.data)
The area under the curve is 0.647.The data was very skewed (90% No,10% Yes) and keeping that in mind I think this is a good model as other techniques like decision trees have not been able to create any split.
Is there any other way to improve this model??
Also,the variables are highly correlated,so would PCA enhance the model performance??