Fuzzy CMAC With incremental Bayesian Ying-Yang learning and dynamic rule construction.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Data Science

Background:

• The Bayesian Ying-Yang (BYY) learning technique, inspired by Taoism, harmonizes data and solutions for clustering.
• Previous applications of BYY learning to fuzzy cerebellar model articulation controllers (FCMAC) were unsuitable for time series data.
• Addressing this limitation, an incremental BYY learning technique is proposed, integrating sliding window and dynamic rule structure algorithms.

Discussion:

• The proposed incremental BYY learning technique incorporates an online expectation-maximization algorithm with a sliding window for efficient fuzzification.
• This approach significantly reduces memory requirements by eliminating the need to process the entire dataset for predictions.
• A rule structure dynamic algorithm dynamically manages rules, mitigating the curse of dimensionality inherent in FCMAC.

Key Insights:

• The incremental BYY learning technique offers a novel solution for time series data analysis.
• The integration of sliding window and dynamic rule algorithms enhances FCMAC's suitability for real-time streaming data.
• Experimental results on currency exchange rates and Mackey-Glass datasets validate the model's effectiveness.

Outlook:

• The developed model shows promise for real-time streaming data analysis applications.
• Further research could explore the application of this technique to other complex time series forecasting problems.
• Optimization of the dynamic rule structure algorithm may lead to further improvements in scalability and efficiency.