An Enhanced Knowledge Salp Swarm Algorithm for Solving the Numerical Optimization and Seed Classification Tasks.

Area of Science:

• Computational Intelligence
• Machine Learning Optimization
• Algorithm Development

Background:

• The standard Salp Swarm Algorithm (SSA) exhibits limitations, including susceptibility to local optima and inadequacy for complex tasks like hyperparameter optimization.
• Seed classification, particularly with Support Vector Machines (SVMs), requires robust optimization techniques to achieve high accuracy.

Purpose of the Study:

• To introduce an Enhanced Knowledge-based Salp Swarm Algorithm (EKSSA) that addresses the limitations of the basic SSA.
• To improve the global and local search capabilities of the Salp Swarm Algorithm.
• To enhance the performance of Support Vector Machines (SVMs) for seed classification through optimized hyperparameter tuning.

Main Methods:

• The EKSSA integrates adaptive parameter adjustment mechanisms (c1 and α) for exploration-exploitation balance.
• A Gaussian walk-based position update strategy is employed to bolster global search efficacy.
• A dynamic mirror learning strategy is introduced to expand the search domain and strengthen local search capabilities.

Main Results:

• The EKSSA demonstrated superior performance over eight state-of-the-art algorithms on thirty-two CEC benchmark functions.
• The EKSSA-SVM hybrid classifier achieved significantly higher classification accuracy in seed classification tasks compared to baseline methods.
• Comparative analysis validated the effectiveness of EKSSA against algorithms like GWO, AOA, and the original SSA.

Conclusions:

• The proposed EKSSA effectively overcomes the local optima problem and enhances both global and local search abilities.
• EKSSA provides a robust optimization framework suitable for machine learning hyperparameter tuning, particularly for SVMs in seed classification.
• The developed EKSSA-SVM hybrid model represents a significant advancement in seed classification accuracy.