Autonomous data partitioning for type-2 fuzzy set based time series

Time series forecasting is widely used to predict future values in several applications, such as climate, industries demand, stock markets and business strategies. Fuzzy set based time series (FTS) forecasting models are widely used to solve forecasting problems because of fuzzy logic’s ability to d...

Full description

Saved in:
Bibliographic Details
Published in:Evolving systems 2024-04, Vol.15 (2), p.575-590
Main Authors: Vargas Pinto, Arthur C., da Silva, Larissa C. C., Silva, Petrônio C. L., Guimarães, Frederico G., de Aguiar, Eduardo P.
Format: Article
Language:English
Subjects:
Accuracy
Algorithms
Artificial Intelligence
Complex Systems
Complexity
Design parameters
Engineering
Forecasting
Fuzzy logic
Fuzzy sets
Mathematical models
Neural networks
Original Paper
Partitioning
Performance evaluation
Root-mean-square errors
Time series
Uncertainty
Citations: Items that this one cites
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Time series forecasting is widely used to predict future values in several applications, such as climate, industries demand, stock markets and business strategies. Fuzzy set based time series (FTS) forecasting models are widely used to solve forecasting problems because of fuzzy logic’s ability to deal with uncertainties. But designing a FTS model can be a complex task since there are important hyperparameters to be defined that directly influence performance, mainly the number of fuzzy sets, model order, and membership function used. Therefore, this paper proposes ADPT2FTS, an interval type-2 FTS model that uses an Autonomous Data Partitioning Algorithm (ADP), capable of improving uncertainty handling and relying on an autonomous method for defining the number of fuzzy sets to be used. We also investigated how changing model order and membership function influenced forecasting accuracy to choose the best parameters to design the best performing model possible. ADPT2FTS performance was evaluated using two complex financial datasets: the European Brent Oil Prices (EBOP) and the Taiwan Capitalization Weighted Stock Index (TAIEX); and two benchmark time series in the literature: University of Alabama Enrollments and Yealy Sunspot. The proposed model was compared to “state-of-art” forecasting models in the literature, designed with several different working mechanisms such as type-1 and type-2 fuzzy logic, neural networks and regression algorithms. Results show that ADPT2FTS achieved better performance and lower error values in terms of RMSE and MAPE metrics.
ISSN:1868-6478
1868-6486
DOI:10.1007/s12530-023-09532-x