Try to use cross validation during the building of your model and include all data sets during the cross validation. In this sense you can evaluate the quality of the data on the obtained r-square from different subsets of your data.
TzeHuey TAM are you using the random forest for a regression task in this situation? Because your validation is better than the training I would suggest going with deeper trees and look what will happen there. Also what variables do you have? Continous, discrete, categorical or mixed? As well you can try to use XGBOOOST, LightGBM or Catboost.
Mantas Lukauskas yupe, i applied random forest for regression. all my variables are continuous. deeper trees means ntree set to higher , eg 1000? I will use the XGBoost, lightGBM and Catboost
TzeHuey TAM for example in scikit-learn random forest function there is parameter "max_depth", also another one min_samples_split that can push tree to be deeper or