Self-Organizing Stacked Type-2 Fuzzy Neural Network With Rule Generalization.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Fuzzy Systems

Background:

• Type-2 fuzzy neural networks (T2FNNs) are powerful tools for nonlinear system modeling.
• Multicollinearity, arising from overlapping Footprint of Uncertainty (FOU), often causes generalization biases in T2FNNs.

Purpose of the Study:

• To introduce a novel self-organizing stacked T2FNN with rule generalization (RG-SOST2FNN).
• To address and mitigate multicollinearity issues in T2FNNs for enhanced performance.

Main Methods:

• A stacked technique using cosine smart priority for T2FNN fusion with sparse, non-collinear inputs.
• A dynamic stacked framework with rule cluster generation for rule adjustment and diversity.
• A stacked risk mitigation algorithm and sparse gradient learning for parameter optimization.

Main Results:

• The RG-SOST2FNN effectively reduces collinearity dependence and parameter estimation variance.
• The proposed method achieves state-of-the-art performance in complex systems, even with high multicollinearity.

Conclusions:

• The RG-SOST2FNN offers a robust solution to multicollinearity in T2FNNs.
• This approach significantly enhances the generalization capability and overall performance of fuzzy neural networks.