Time series; Ensembles; Fuzzy rules

There is no individual forecasting method that is generally for any given time series better than any other method. Thus, no matter the efficiency of a chosen method, there always exists a danger that for a given time series the chosen method is inappropriate. To overcome such a problem and avoid the above mentioned danger, distinct ensemble techniques that combine more individual forecasting methods are designed. These techniques basically construct a forecast as a linear combination of forecasts by individual methods. In this contribution, we construct a novel ensemble technique that determines the weights based on time series features. The protocol that carries a knowledge how to combine the individual forecasts is a fuzzy rule base (linguistic description). An exhaustive experimental justification is provided. The suggested ensemble approach based on fuzzy rules demonstrates both, lower forecasting error and higher robustness.