I want to train an ML model with 3 input parameters 1 output using Xgboost regressor. My output data is time-dependent meaning that I have the values of output parameters at 500 time steps. Also, I have 1000 measurements; this means that I have 1000 input values and 1000 output values data at each time step. Now, I can train my model in two ways:
- Train an ML model for each time step separately. I have to do the training 500 times (for all time steps) and I have 1000 input and output values at each time step.
- Considering time as an additional input to the problem, and train the whole model once for all time steps. This way I have 500000 (500*1000) input and output values and I need to train the ML model once.
I am not sure which method is the best. Any suggestions in this regard? Do they have any preferences in comparison to each other at all? Any reference where I can read more about this?