Multi-Swarm Algorithm for Extreme Learning Machine Optimization.

Area of Science:

• Machine Learning
• Computational Intelligence
• Optimization Algorithms

Background:

• Extreme Learning Machines (ELMs) are fast and efficient but their performance relies heavily on hidden layer weights and biases.
• Optimizing these parameters is an NP-hard problem, limiting ELM's practical application.
• Existing methods struggle to find optimal parameters, hindering ELM's full potential.

Purpose of the Study:

• To develop an effective method for determining optimal or near-optimal weights and biases for ELMs.
• To address the challenge of parameter optimization in ELMs for improved performance.
• To propose a novel multi-swarm hybrid optimization approach for ELM tuning.

Main Methods:

• A multi-swarm hybrid optimization approach combining Artificial Bee Colony (ABC), Firefly Algorithm (FA), and Sine-Cosine Algorithm (SCA).
• Validation on seven benchmark classification datasets.
• Comparison with existing state-of-the-art approaches.

Main Results:

• The proposed multi-swarm hybrid optimization approach achieved superior generalization performance compared to other methods.
• Demonstrated improvements in accuracy, precision, recall, and F1-score.
• Further experiments confirmed the superiority of the three-algorithm hybrid over two-algorithm combinations.

Conclusions:

• The proposed multi-swarm hybrid optimization technique effectively optimizes ELM parameters, leading to enhanced classification performance.
• Combining three swarm intelligence algorithms (ABC, FA, SCA) yields better results than pairwise combinations.
• The developed ELM tuning framework offers a promising solution for practical applications requiring rapid decision-making models.