A Forecasting Model Based on High-Order Fluctuation Trends and Information Entropy.

Area of Science:

• Financial forecasting
• Time series analysis
• Computational intelligence

Background:

• Existing high-order prediction models often overlook historical data inconsistencies and fluctuation trends.
• These overlooked characteristics are crucial for accurately describing and predicting historical market behavior.
• Traditional models struggle with rule generation and lack generality when dealing with complex data patterns.

Purpose of the Study:

• To propose a novel model for time series forecasting that incorporates historical fluctuation trends and inconsistencies.
• To extend fuzzy logical relationships (FLRs) to neutrosophic logical relationships (NLRs) for enhanced prediction.
• To improve the generality and accuracy of high-order prediction models.

Main Methods:

• Logical rules are trained by mapping dynamic trend states (up/down) to truth/false-membership in neutrosophic sets.
• Information entropy quantifies historical inconsistency, mapped to the indeterminacy-membership of neutrosophic sets.
• Forecasting utilizes neutrosophic set similarities to identify similar historical patterns for prediction.

Main Results:

• The proposed neutrosophic logical relationship (NLR) model demonstrates stable prediction ability across different datasets.
• Experimental results show superior prediction accuracy compared to traditional discrete high-order fuzzy logical relationships (FLRs).
• The model effectively handles the problem of insufficient rules in traditional prediction methods.

Conclusions:

• The developed NLR model offers higher generality and robustness in time series forecasting.
• Incorporating fluctuation trends and inconsistencies significantly enhances prediction accuracy and universality.
• The model proves effective for forecasting stock market indices like the Taiwan Stock Exchange and Heng Seng Index.