When you test several values for a parameter (for example 100 and 300 trees for a Random Forrest), a cross validation is done on TRAIN dataset to choose the best value.

• By defaut, a random K-Folds is applied, which is suitable for a Random TRAIN/TEST split, BUT NOT for a Time split.
  
  
• For a Time split for the TRAIN and TEST datasets, you must applied the same strategy for the cross validation.
  
  
  • Select "Custom code"
    
  
  
  • and choose "Time series cross validation" in the available "{} Code Samples"