Abstract: Methods, computer program products, and systems are presented. The methods include, for instance: generating a drift annealed time series prediction model based on training data. In one embodiment the generating may include: recording an ensemble of candidate models for at least one predictor variable of the training data responsive to creating the ensemble based on the training data by machine learning. The ensemble includes three candidate models represented by respective prediction function to formulate a potentially predictive relationship between a target variable and predictor variables in the training data. Respective candidate models in the ensemble is manipulated to adjust degrees associated with predictor variables such that respective new models take relative importance of predictor variables into account and the drift annealed time series prediction model based on the new models is produced.