Fuzzy Broad Learning System: A Novel Neuro-Fuzzy Model for Regression and Classification

Area of Science:

• Artificial Intelligence
• Machine Learning
• Computational Intelligence

Background:

• Broad Learning System (BLS) offers rapid learning but can be limited in handling complex, nonlinear data.
• Neuro-fuzzy systems combine the strengths of neural networks and fuzzy logic for robust decision-making.
• Existing neuro-fuzzy models often suffer from the 'rule explosion' problem, increasing computational complexity.

Purpose of the Study:

• To propose a novel neuro-fuzzy model, the fuzzy Broad Learning System (BLS), by integrating Takagi-Sugeno (TS) fuzzy systems into BLS.
• To enhance the feature representation and nonlinear transformation capabilities within the BLS framework.
• To evaluate the performance and efficiency of the proposed fuzzy BLS against existing state-of-the-art models.

Main Methods:

• The fuzzy BLS replaces standard feature nodes with multiple TS fuzzy subsystems for parallel data processing.
• An enhancement layer is introduced for nonlinear transformation of fuzzy subsystem outputs before final aggregation.
• The k-means method is used for determining Gaussian membership function centers and the number of fuzzy rules.

Main Results:

• The fuzzy BLS demonstrated superior performance in benchmark regression and classification tasks compared to other non-fuzzy and neuro-fuzzy models.
• The proposed model achieved faster training times and required fewer fuzzy rules than traditional neuro-fuzzy approaches.
• Analytical calculation of model parameters ensures the fast computational efficiency characteristic of BLS.

Conclusions:

• The fuzzy BLS effectively merges TS fuzzy systems with BLS, offering improved accuracy and efficiency.
• This novel approach mitigates the rule explosion problem common in neuro-fuzzy systems.
• Fuzzy BLS presents a promising alternative for complex machine learning tasks requiring both speed and high performance.