How to perform backward chaining in logistic regression using R



I am currently solving one classification problem using logistic regression algorithm in R .I have created the model using step function, but my step function only performs forward chaining.I want to know is there a way by which I can perform backward chaining.

mod<-step(glm(Dependent~., family = binomial(link=logit),data = train_final))

I also want to know what does AIC means in the model and how it values affect the model.



set direction=c(“backward”) in step , check the documentation as well :slight_smile: if you want a more robust methods check in the package “leaps” and the function regsubsets() as you do have many variables it could find out quickly the most important variables.
AIC is one metrics, for parametric model, it is base on information loss check wikipedia it is well explained there, in few words the AIC compare two models and base on this metric which is dependant on the number of variables in the models calculate a probability of information loss.
If you use regsubsets() you could have AIC as well as R2 of course and BIC, which could be handy and easier to interpret. As you use a sigmoid you will no see a lot a variation.
Hope this help.