An Improved Ensemble of Random Vector Functional Link Networks Based on Particle Swarm Optimization with Double...

Area of Science:

• Machine Learning
• Ensemble Methods
• Artificial Neural Networks

Background:

• Ensemble learning performance hinges on classifier selection and combination.
• Random Vector Functional Link Networks (RVFL) offer fast learning, simple structure, and good generalization.

Purpose of the Study:

• To propose an improved RVFL ensemble using a double optimization strategy with ARPSO.
• To achieve a more compact ensemble with enhanced convergence performance.

Main Methods:

• Utilizing ARPSO for selecting optimal base RVFLs based on convergence accuracy and diversity.
• Initializing ensemble weights with minimum norm least-square and optimizing with ARPSO.
• Pruning redundant RVFLs to create a compact ensemble.

Main Results:

• The proposed ARPSO-based RVFL ensemble demonstrates superior performance compared to single optimization methods.
• Experimental results confirm improved convergence accuracy and reduced system complexity.
• A practical strategy for pruning redundant base classifiers is presented for classification and regression.

Conclusions:

• The double optimization strategy with ARPSO effectively enhances RVFL ensemble performance.
• The proposed method yields more compact and accurate ensemble systems.
• The developed pruning strategy offers a feasible approach for simplifying ensemble models.