Boosting ridge for the extreme learning machine globally optimised for classification and regression problems.

Area of Science:

• Machine Learning
• Ensemble Methods
• Computational Intelligence

Background:

• Extreme Learning Machines (ELM) utilize preconfigured hidden layers with output layer optimization.
• Existing Boosting Ridge ELM (BRELM) implementations train base learners sequentially, which can lead to saturation.
• Traditional ensemble methods often rely on strong classifiers, which may not always be optimal.

Purpose of the Study:

• To propose a novel global learning method for the Boosting Ridge (BR) framework using Extreme Learning Machines (ELM).
• To address the limitations of sequential training in BRELM by optimizing all base learners simultaneously.
• To improve ensemble performance by allowing diverse hidden layer configurations and avoiding reliance on strong classifiers.

Main Methods:

• A global ensemble learning approach is proposed for the BR framework.
• Base learners are optimized concurrently in a single step, rather than sequentially.
• Each base learner features distinct hidden layer configurations.

Main Results:

• The proposed global ensemble method demonstrates superior performance compared to the original BRELM implementation.
• Statistical tests confirm the method's effectiveness across various regression and classification benchmark datasets.
• The approach shows significant improvements regardless of dataset characteristics like size, class number, or imbalance.

Conclusions:

• The novel global learning method within the BR-ELM framework offers enhanced performance for both regression and classification.
• Simultaneous optimization of diverse base learners overcomes limitations of sequential training and strong classifier dependency.
• This approach represents a significant advancement in ensemble learning for ELM.