Multi-variate timeseries forecasting using complex fuzzy logic

Complex fuzzy logic has been repeatedly used to construct very effective time-series forecasting algorithms. The great majority of these studies, however, only involve univariate time series. The only exception is one work on bivariate time series. Our objective is to investigate the network architectures and time series representations that lead to effective general multi-variate time series forecasting. Our experiments will make use of the Adaptive Neuro-Complex Fuzzy Inferential System architecture, evaluating three different approaches (single-input single-output, multiple-input single-output, and multiple-input multiple-output) on three multi-variate datasets. Our results indicate that the complex fuzzy architectures are at least as accurate as Radial Basis Function Networks and Support Vector Regression on these problems.