So it's not just me! The stuff people do in Kaggle competitions is mind blowing. They do models within models within models, do weird interaction terms with their features and transform them with different functions, although most of that is to go from 94% accuracy to 97% accuracy...