Título: Dynamic fuzzy systems for modelling non-stationary time series
Autores: Luna, Ivette
Resumo: Time series modelling is a subject of interest to several fields of knowledge. The challenge of dealing with this type of data is to develop computational instrumentals able to handle with different features related to non-stationarity as well as uncertainty present in most of real processes. In that context, dynamic fuzzy systems have shown capable of dealing with these aspects. Hence, this paper presents a case study that aims to provide empirical evidence that validate the ability of dynamic fuzzy systems for modelling and forecasting non-stationary time series. The dynamic fuzzy model is based on Takagi-Sugeno fuzzy systems, with a learning algorithm based on the recursive version of the Expectation Maximization optimization method. The study considers the modelling of a bond price time series. The model is evaluated with and without the dynamic learning in order to verify the effect of the learning process over the model performance. Additionally, the fuzzy model is also compared to an offline neural network. The results show the potential of fuzzy systems with dynamic learning for modelling non-stationary time series and changes over time.
Código DOI: 10.21528/CBIC2013-138
Artigo em pdf: bricsccicbic2013_submission_138.pdf
Arquivo BibTex: bricsccicbic2013_submission_138.bib