A Multi-Strategy Enhanced Ivy Algorithm for Optimizing GANs to Improve Imbalanced Data Classification

The Ivy Algorithm (IVYA), an emerging swarm intelligence optimization method, often faces challenges of slow convergence, local optima entrapment, and an imbalanced exploration-exploitation trade-off when applied to complex, high-dimensional problems. To address these deficiencies, this paper proposes an Enhanced Ivy Algorithm (E-IVYA). E-IVYA integrates three synergistic strategies: an elite-guided opposition-based learning mechanism to enhance population diversity, a stagnation response strategy to effectively escape local optima, and an adaptive movement strategy inspired by the Sine-Cosine Algorithm to dynamically balance global exploration and local exploitation. Experimental validation on the challenging IEEE CEC 2014 and 2017 benchmark suites demonstrates that E-IVYA’s performance significantly surpasses that of the original IVYA and various classic and advanced algorithms, including PSO, GWO, and LSHADE. Furthermore, E-IVYA shows excellent practical potential by optimizing the hyperparameters of Generative Adversarial Networks (GANs), which substantially improves classification performance on imbalanced datasets. These findings establish E-IVYA as a robust and superior optimizer, successfully overcoming the inherent limitations of the original algorithm.