Performance improvement of classifier fusion for batch samples based on upper integral.

Area of Science:

• Machine Learning
• Artificial Intelligence
• Computational Mathematics

Background:

• Extreme Learning Machines (ELMs) are effective but can suffer from generalization issues.
• Existing classifier fusion methods, like fuzzy integrals, have limitations.
• Improving the generalization ability of ELMs is crucial for robust classification.

Purpose of the Study:

• To propose a novel scheme for fusing ELMs using upper integrals.
• To enhance the classification efficiency and accuracy of ELMs.
• To provide a theoretical guarantee that the fused ELM's accuracy is superior to individual ELMs.

Main Methods:

• Developed a new ELM fusion scheme based on upper integrals.
• Utilized upper integrals to assign samples to different ELMs for optimal classification.
• Solved an optimization problem involving upper integrals to determine sample assignment proportions.
• Compared the proposed method with existing fusion techniques like Bagging and Boosting.

Main Results:

• The proposed upper integral-based fusion scheme improves ELM generalization ability.
• Theoretical analysis confirms the fused ELM's accuracy is not less than any individual ELM.
• Numerical simulations show improvements over existing methods like Bagging and Boosting.
• The method effectively assigns samples to maximize classification efficiency.

Conclusions:

• The upper integral-based ELM fusion offers a theoretically sound and practically effective approach.
• This method provides a significant advancement over existing fuzzy integral models for classifier fusion.
• The proposed technique enhances classification performance and efficiency in machine learning applications.