I've got data-set which is very large (16 million + ) and has over 150 covariates (which some are hot-encoded). This data-set spans across roughly 10 years These covariate inlcude details of the policy holders. My objective are to determine the probability of a large claim from a given policy holder and the predict the severity of a large claim (Cost of the paid out) for a given policy holder. However large claims are sparse and there are roughly 500 in a data-set. Most of the data-set consist of no claims and roughly 50000 are normal claims. The defintion of a large claim is something which over £90,000 to paid out(However this can be redefined for Peak over Threshold if required). My automatic thought is to use extreme value theory however I'm not sure how to approach it. It would be great if you could give step by step plan of how to approach this. (preferable in Python or R)
I've only dealt with univariate extreme value theory before, so my knowlegde is quite limited. It would be ideal to include some significant covariates in the model.
Any help would be much appreciated.