Adaptive time-variant models for fuzzy-time-series forecasting.

Area of Science:

• Computational intelligence
• Time series analysis
• Fuzzy logic systems

Background:

• Fuzzy time series models are used for predictions in diverse fields like enrollment and stock markets.
• Existing models often rely on fixed analysis window sizes, potentially limiting prediction accuracy.
• Key factors influencing fuzzy time series forecasting include discourse partition, rule content, and defuzzification methods.

Purpose of the Study:

• To propose an adaptive time-variant fuzzy-time-series forecasting model (ATVF) for enhanced prediction accuracy.
• To develop a model that automatically adjusts the analysis window size based on training phase accuracy.
• To improve upon existing fuzzy time series forecasting techniques.

Main Methods:

• The proposed ATVF model dynamically adapts its analysis window size during the training phase.
• Heuristic rules are employed in the testing phase to generate forecasting values.
• The model's performance is evaluated using both simulated and real-world time series data.

Main Results:

• The ATVF model demonstrated significant improvements in forecasting accuracy compared to traditional fuzzy time series models.
• Testing on University of Alabama enrollments and the Taiwan Stock Exchange Capitalization Weighted Stock Index (TAIEX) validated the model's effectiveness.
• The adaptive nature of the ATVF model contributes to its superior performance.

Conclusions:

• The adaptive time-variant fuzzy-time-series forecasting model (ATVF) offers a more accurate approach to time series prediction.
• Dynamic adjustment of the analysis window size is a key factor in improving forecasting performance.
• The ATVF model provides a robust alternative for applications requiring high-accuracy time series forecasting.