In Model selection piece, use Bayesian Optimization to do hyper-parameter tuning
Look at adding a double lift chart, this will also indicate model diversity for possible ensembling
Or for classification look at confusion matrix and see if two models are doing well on different segments