collect the results and sort them so you could easily see which models work best. For Target 0/1 models Gini coefficient and Top Decile Lift should be your guide (the greater the better). For continous / numeric Targets you could use RMSE (0 is best) or Spearman if you just want a really good correlation between your score and the real values (but not necessary exactly predict them). please download the complete DeepLearning (Keras, Tensorflow, H2O.ai) Workflow group: https://kni.me/s/sl4rYEzjFPm7a99a