A Type-2 Fuzzy Time Series Classification System with Optimized Time Period Selection
Ashish Bhatia, Hani Hagras · 2024
Time-series analysis plays a crucial role in numerous real-world applications, ranging from financial forecasting to environmental monitoring and beyond. Traditional classification techniques often struggle to effectively handle the uncertainties and imprecision inherent in time series data. To address this challenge, fuzzy time series models have emerged as a promising alternative, offering a flexible framework capable of capturing the uncertainties and vagueness intrinsic to temporal patterns. In this paper, we propose an Optimized Time Series Interval Valued Fuzzy System (OTS-IVFS) that leverages the power of type-2 fuzzy logic to handle temporal data effectively and provide fully explainable models. We have performed experiments on five data sets from a diverse set of use cases. Our system significantly outperforms the best-in-class algorithms for the Earthquake dataset by 16.67% increase in accuracy while giving comparable results in 3 other datasets, whilst maintaining full interpretability.