Transformer and Variational Autoencoder Hybrid Models Optimized by Nature-Inspired Metaheuristics for Diabetes Diagnosis and Risk Assessment: A Systematic Review (۲۰۱۷–۲۰۲۵)

The rapid increase in diabetes prevalence worldwide has created an urgent need for accurate, automated diagnostic systems capable of handling complex, high-dimensional clinical data. Recent advances in deep learning, particularly Transformer-based architectures and Variational Autoencoders (VAEs), have demonstrated superior capability in modeling long-range dependencies and learning robust latent representations of tabular and sequential medical data. However, the performance of these models heavily depends on optimal hyperparameter configuration and architectural design, which remain challenging in noisy, imbalanced diabetes datasets. Nature-inspired metaheuristic algorithms such as Harris Hawks Optimization (HHO), Quantum Firefly, and their variants have emerged as powerful tools for automated tuning of these sophisticated models. This systematic literature review (SLR), covering studies from ۲۰۱۷ to ۲۰۲۵, analyzes ۳۰ high-quality papers that propose hybrid frameworks combining Transformers and VAEs optimized via modern metaheuristics for diabetes detection, risk stratification, and glucose dynamics forecasting. We present a novel taxonomy, detailed comparative analysis, and evidence-based performance benchmarks showing that metaheuristic-optimized hybrids consistently outperform standalone deep models by ۵–۱۵% in accuracy and AUC while significantly reducing false positives—a critical requirement in clinical decision-making. The review concludes with open challenges and a research roadmap emphasizing multi-modal integration, edge deployment, and explainable AI for real-world clinical adoption.

کلیدواژه ها:

Transformer ، Variational Autoencoder ، Nature-Inspired Metaheuristics ، Diabetes Diagnosis ، Harris Hawks Optimization (HHO)