Improving atmospheric angular momentum forecasts by machine learning