<p>Missing data imputation is a critical task, as incomplete datasets can negatively impact model performance. Traditional imputation methods often fail to capture the intricate relationships within data, resulting in poor representations of complex patterns and reduced predictive accuracy. Generative Adversarial Network (GAN)-based imputation methods frequently encounter challenges, such as hyperparameter tuning difficulties and mode collapse. In this paper, we propose a improved GAN-based imputation framework, termed BESGAIN. To be concrete, BESGAIN introduces the Bald Eagle Search (BES) optimization algorithm to dynamically adjust hyperparameters, ensuring high-quality data imputation. Additionally, we incorporate an enhanced loss function that leverages cosine similarity to improve data diversity. This helps prevent performance degradation caused by overfitting to a single data mode. As a result, BESGAIN achieves a richer and more accurate set of imputed values. We validate the performance of BESGAIN across seven real-world datasets and compare it against state-of-the-art and widely used imputation methods. The experimental results demonstrate that BESGAIN outperforms existing techniques in terms of accuracy, robustness, and generalizability, particularly in datasets with high missing rates.</p>