Best approach:

1) Train different models with different parameters

2) determine an evaluation metric (example: MSE)

3) for different model, draw graph (scatterplot / line plot)

    - model on X-axis

    - MSE value on Y-axis

4) the model with the lowest MSE is the best model