Subsethood-product fuzzy neural inference system (SuPFuNIS).

Area of Science:

• Computational Intelligence
• Artificial Neural Networks
• Fuzzy Logic Systems

Background:

• Existing fuzzy systems often struggle to integrate diverse data types seamlessly.
• The need for adaptable and robust inference systems in complex domains is growing.

Purpose of the Study:

• To introduce a novel Subsethood-Product Fuzzy Neural Inference System (SuPFuNIS).
• To demonstrate SuPFuNIS's capability to process both numeric and linguistic inputs concurrently.
• To evaluate SuPFuNIS performance across diverse benchmark problems.

Main Methods:

• Development of SuPFuNIS featuring tunable input fuzzifiers and Gaussian fuzzy set connections.
• Utilized fuzzy mutual subsethood-based activation spread and product operator for rule firing.
• Employed supervised gradient descent for training and a volume-defuzzification process for output.

Main Results:

• SuPFuNIS successfully integrated numeric and linguistic data, demonstrating high flexibility.
• The system achieved excellent performance in time series prediction, data classification, medical diagnosis, and function approximation.
• Expert knowledge integration was effectively demonstrated using a truck backer-upper control problem.

Conclusions:

• SuPFuNIS offers a powerful and versatile approach to fuzzy inference.
• The model's architecture allows direct translation of rule-based knowledge.
• SuPFuNIS shows significant potential for various real-world applications, outperforming existing models.