Hybrid whale algorithm with evolutionary strategies and filtering for high-dimensional optimization: Application to...

Area of Science:

• Computational Intelligence
• Optimization Algorithms
• Machine Learning

Background:

• The standard whale optimization algorithm (WOA) suffers from suboptimal results and inefficiencies, particularly in high-dimensional spaces.
• Initial population generation in WOA often leads to uneven distribution and low diversity, hindering performance.
• Support Vector Machines (SVM) are effective for high-dimensional data but benefit significantly from parameter optimization.

Purpose of the Study:

• To propose a novel optimization algorithm, the Recombinant Evolutionary Strategy-enhanced Whale Optimization Algorithm (RESHWOA), by fusing WOA with a discrete recombinant evolutionary strategy.
• To enhance the initialization diversity of the whale optimization algorithm.
• To apply the proposed RESHWOA for optimizing Support Vector Machine (SVM) parameters on high-dimensional microarray cancer datasets.

Main Methods:

• Fusion of the standard whale optimization algorithm (WOA) with a discrete recombinant evolutionary strategy to create RESHWOA.
• Comparative simulation experiments on thirteen benchmark test functions (unimodal and multimodal) against the original WOA.
• Application of RESHWOA and WOA for optimizing SVM parameters on six microarray cancer datasets, utilizing Bhattacharya distance and signal-to-noise ratio for data reduction.

Main Results:

• RESHWOA demonstrated superior performance over the standard WOA on benchmark functions, showing improvements in accuracy, minimum mean, and reduced standard deviation.
• The proposed RESHWOA effectively addressed the shortcomings of the original WOA, particularly concerning initialization diversity and convergence.
• Optimization of SVM parameters using RESHWOA on microarray datasets yielded enhanced performance compared to WOA.

Conclusions:

• The developed RESHWOA significantly outperforms the standard WOA in addressing optimization challenges in high-dimensional spaces.
• The fusion strategy effectively enhances initialization diversity, leading to more robust and accurate optimization results.
• RESHWOA provides a powerful tool for optimizing machine learning models like SVM, especially for complex biological datasets.