Multiscale Wavelet Preprocessing for Fuzzy Systems

This paper examines the prediction performance of fuzzy subtractive-clustering models on time series with trends, seasonalities, and discontinuities and indicates that wavelet preprocessing improves forecast accuracy for time series that exhibit variance changes and other complex local behavior.

Abstract

Fuzzy systems, also referred to as universal approximators, have been used to model real-world data. In this paper, we examine the prediction performance of fuzzy subtractive-clustering models on time series with trends, seasonalities, and discontinuities. Our results indicate that wavelet preprocessing improves forecast accuracy for time series that exhibit variance changes and other complex local behavior. Conversely, for time series that exhibit no significant structural breaks or variance changes, fuzzy models trained on raw data perform better than hybrid fuzzy-wavelet models. Further work is required to investigate the use of wavelet variance profile of time series to determine the suitability of the application of wavelet-based preprocessing on prediction models.