I'm considering the benefits of building Elastic Net models over GLM for claims severity. Can someone advise whether EN models are more robust to outliers/large values compared to GLM, and why?
That's a little beyond the scope of SP8 Amelhypo, but my understanding is that Elastic Net Regularisation can be used in conjunction with GLMs, rather than instead of GLMs. It's a technique that helps you remove highly correlated predictors that don't improve the model accuracy. There's plenty of literature available online if you search for it but, since it's not covered in our Acted products, I'll leave other members of this forum to post their replies ...
