Epsilon-insensitive fuzzy c-regression models: introduction to epsilon-insensitive fuzzy modeling

Area of Science:

• Fuzzy logic
• Machine learning
• Regression analysis

Background:

• Traditional fuzzy modeling often uses crisp loss functions, leading to inconsistencies when fitting models to real-world data.
• Existing methods may struggle with outlier robustness and controlling generalization ability.

Purpose of the Study:

• Introduce a novel epsilon-insensitive fuzzy c-regression model (epsilonFCRM) for improved fuzzy modeling.
• Address the intrinsic inconsistency of using crisp loss functions in fuzzy modeling.
• Enhance robustness to outliers and provide better control over generalization ability.

Main Methods:

• Utilizes a weighted epsilon-insensitive loss function for fitting fuzzy regression models to data.
• Develops a method based on human cognitive processes for fuzzy modeling.
• Formulates the problem as multiple simultaneous quadratic programming problems.

Main Results:

• The proposed epsilonFCRM method effectively fits real data while mitigating inconsistencies.
• Demonstrates improved robustness against outliers and controllable generalization.
• The incremental learning method provides an efficient numerical solution.

Conclusions:

• The epsilon-insensitive fuzzy modeling approach offers a more intuitive and effective way to build fuzzy models.
• The method successfully addresses limitations of traditional fuzzy regression techniques.
• Validated through examples, showcasing its practical applicability in fuzzy modeling.