Add best RF model predictions to features dataframe

[jds485]

We want to try using the RF model predictions as input to LSTM and RGCN models to see if that improves performance. Can try:

• best RF model with only static features
• best overall RF model with static and dynamic features