Time series forecasting using fuzzy techniques

The aim of this contribution is to show the opportunities of applying of fuzzy time series models to predict multiple heterogeneous time series, given at International Time Series Forecasting Competition [http://irafm.osu.cz/cif/main.php]. The dataset of this competition includes 91 time series of different length, time frequencies and behaviour. In this paper the framework (algorithm) of multiple time series forecasting, based on fuzzy techniques, is proposed. We applied the traditional decomposition of time series, using F-transform technique. To identify the model of the time series three fuzzy time series models were tested: model, based on fuzzified time series values, model, based on fuzzified first differences of time series values and model, based on the fuzzy tendency. To choose the best model we introduce two step algorithm and new criteria in addition to well-known, based on linguistic description of time series fuzzy tendency. In the conclusion the received results are discussed and the efficiency of the proposed approach is shown.