Online Ensembles for Time Series Forecasting

Online Ensembles for Time Series Forecasting

Usually online learning refers to a form of learning where an algorithm has access to the loss of a prediction after each prediction, in which case it can adjust itself for subsequent predictions. In some cases online learning is a more natural description of learning problems and has been well studied, however, not many implementations exist for online learning for ensemble based time series forecasting. The following is a project I worked on that describes implementing these types of learning algorithms as ensemble algorithms for baskets of various time series forecasting methods. Paper; Code.