So I am new in R Project and I want to run ROC on my Ebola dataset for optimal model and robustness selection. Its a predictive modeling of ebola patients clinical and non clinical characteristics and their treatment outcome.
1. Can I use ROC to select my optimal model?
2. And how can I interprete my ROC results in terms of sensitivity, specificity and robustness?
Is this a logistic regression model you are talking about? I strongly suggest you have a look at Frank Harrell's book Regression Modeling Strategies for how to choose the optimal model. You should also consider how you intend to validate the model (normally you would do an internal validation first).
The AUC (or c-statistic) of an ROC is just one of the discrimination metrics of a model. You need to consider calibration and possibly other factors as well. The ROC is simply a plot of sensitivity verse 1-specificity so won't give you a simple single sensitivity or specificity.
A good starting point for this problem is "An introduction to statistical learning with applications in R" book which is freely availabel here:
http://www-bcf.usc.edu/~gareth/ISL/
regards,
Bogdan
Thanks. I am doing both Poisson and Logistic regression. I am curious about how robust is Rproject ROC model selection could be. Yes I am considering cross validation.
- Badges
- Science topic
- Similar topics
- Linguistics
- Composition Studies
- Writing