Evolving granular neural networks from fuzzy data streams

Area of Science:

• Computational Intelligence
• Machine Learning
• Fuzzy Systems

Background:

• Nonstationary environments pose challenges for traditional modeling techniques.
• Fuzzy data streams require adaptive and interpretable modeling approaches.
• Existing methods may struggle with gradual and abrupt parameter changes.

Purpose of the Study:

• Introduce an evolving granular neural network (eGNN) framework.
• Enable fuzzy system modeling from fuzzy data streams.
• Address challenges of nonstationary environments in online learning.

Main Methods:

• Utilize fuzzy neurons for information fusion and interpretable local models.
• Employ an online incremental learning algorithm for neural network structure development.
• Focus on trapezoidal fuzzy intervals and various data types (triangular, interval, numeric).

Main Results:

• Demonstrate successful handling of fuzzy data streams.
• Achieve superior performance in classification and function approximation tasks.
• Outperform state-of-the-art approaches in accuracy, transparency, and compactness.

Conclusions:

• The eGNN framework offers a robust solution for evolving fuzzy system modeling.
• eGNN provides interpretable and compact fuzzy models for complex data.
• The approach is effective for applications in material and biomedical engineering.