MetaSegan: generative AI-driven data augmentation with metaheuristic optimisation for employee turnover prediction

This research proposes a Metaheuristic Self-Attention Generative Adversarial Network (MetaSeGAN) framework to enhance predictive performance in employee attrition analysis within the context of Human Resource Management (HRM). The MetaSeGAN framework leverages a Conditional Tabular GAN to generate high-quality synthetic tabular data, addressing imbalanced datasets and improving training effectiveness. Furthermore, a self-attention mechanism is integrated into the generator network to capture complex dependencies and relationships within the data, allowing the model to focus on key features that significantly influence employee attrition. Experiments conducted on the Kaggle HR Analytics and New England College of Business datasets demonstrate strong performance, achieving accuracies of 99.10% and 88.80%, respectively, and outperforming baseline methods, including CTGAN, SMOTE, ADASYN, and ForestDiffusionGenerator.