I used extensively the theta mention in a forecasting course and it's amazing how good it is, even with default parameters.

ML techniques have negligible improvement, considering the computational overhead that they require.