Evaluation of multivariate transductive neuro-fuzzy inference system for multivariate time-series analysis and modelling

Abstract

Multivariate Transductive Neuro-Fuzzy Inference System model, named the mTNFI is a previously proposed conceptual transductive approach designed for analysis and modelling of multivariate time-series data. In this study, we revisit, implement and evaluate the mTNFI model potential for patterns of relationship extraction from a collection of interrelated time-series data. Results of conducted assessment confirm the mTNFI capability in recognizing patterns of relationship and then to model them in human-readable form, i.e. fuzzy rules. Additionally, a comparative analysis also show the superiority of the mTNFI model in comparison to other widely known time-series forecasting techniques when being used to predict future values.