An Incremental Type-2 Meta-Cognitive Extreme Learning Machine.

Area of Science:

• Machine Learning
• Artificial Intelligence
• Computational Intelligence

Background:

• Existing extreme learning machine (ELM) algorithms struggle with complexity, uncertainty, concept drift, and high dimensionality.
• There is a need for adaptive learning models that can handle dynamic data streams effectively.

Purpose of the Study:

• To propose a novel incremental type-2 meta-cognitive extreme learning machine (eT2ELM) to overcome limitations of existing ELMs.
• To enhance prediction accuracy and model robustness in the presence of concept drift and uncertainty.

Main Methods:

• The eT2ELM incorporates human meta-cognition principles: what-to-learn, how-to-learn, and when-to-learn.
• A certainty-based active learning method is used for sample selection, enabling semi-supervised classification.
• A generalized interval type-2 fuzzy neural network with interval type-2 multivariate Gaussian functions and Chebyshev series is employed.

Main Results:

• The eT2ELM demonstrated superior performance across 12 data streams with varying concept drifts.
• Numerical results were validated through rigorous statistical tests, confirming its efficacy.
• The model achieved reliable prediction while maintaining low computational complexity.

Conclusions:

• The proposed eT2ELM effectively addresses key challenges in machine learning, including complexity and concept drift.
• The meta-cognitive framework enhances the adaptability and performance of extreme learning machines.
• eT2ELM offers a promising solution for real-time, adaptive learning from data streams.