Abstract: A new method is developed to adjust the forecast in case of a small number of observations or in the absence of standard patterns which can be detected by classical machine learning and ...