A synergistic enhancement of the Ivy algorithm for GAN-based imbalanced classification

Area of Science:

• Computational Intelligence
• Swarm Intelligence Algorithms
• Optimization Techniques

Background:

• The standard Ivy Algorithm (IVYA) is a swarm intelligence optimization method inspired by plant growth.
• Balancing exploration and exploitation is critical for IVYA's efficiency in high-dimensional problems.
• Existing methods face challenges in maintaining diversity and escaping local optima.

Purpose of the Study:

• To propose an Enhanced Ivy Algorithm (E-IVYA) to improve optimization performance.
• To address the exploration-exploitation balance challenge in swarm intelligence.
• To enhance the efficiency of automated machine learning, particularly hyperparameter optimization.

Main Methods:

• Introduced a dynamic perturbation framework for population diversity (symmetric and asymmetric exploration).
• Implemented a dynamic escape mechanism using elite differential mutation to avoid stagnation.
• Integrated an adaptive movement strategy inspired by the Sine-Cosine Algorithm for exploration-exploitation balance.

Main Results:

• E-IVYA demonstrated superior performance against benchmark algorithms on IEEE CEC 2014 and 2017 test suites.
• The algorithm achieved a high F1-Score of 0.87 for minority class classification on the Credit-Card Fraud dataset.
• E-IVYA significantly outperformed standard techniques like SMOTE in optimizing Generative Adversarial Networks for imbalanced data.

Conclusions:

• The Enhanced Ivy Algorithm (E-IVYA) is a robust and efficient optimization tool.
• E-IVYA effectively tackles complex, high-dimensional optimization problems.
• The proposed enhancements provide a superior approach for automated machine learning tasks, especially with imbalanced datasets.