Multiple class prediction with unbalanced data

Question

I have 4 different web sites (A/B/C/D) and need to predict whether a user will rather convert on A/B/C or D or not convert at all. My data set has 100000 rows and only 100 of those are converted user (A/B/C), so very unbalanced. My explanatory variables are user session length, device type, refferal ect (most of them are factors). In case I had only one web site it would be quite easy, because I could have used logistic regression. Which statistical technique would you recommend for this type of question.