# 9th place solution

Hello,
It was my first competition here in kaggle. I discovered most of the techniques that I use a few weeks before the competition so I can't complain about the ranking x) (I'm even surprised).
Regarding my solution, I did feature engineering,
MixedPCA
on some specific features, created hundreds of dummies to model the so-called ClaimDescription variable based on word frequency (I did a large part manually : selecting words, combination of words and so on), I also saw that there were patterns at the end of some sentences but had trouble modeling them. I then tried a few boosting methods (
GBM
and
Xgboost
mainly) using the h2o package and neural network models and a belnding after that.
I also learned a lot from the competition public notebooks and tried them after changing their features engineering part.
Happy to found such an amazing community of actuaries here in kaggle ! 😃