Ecosystems: A Comprehensive Review of Transformer Architectures, Quantum Machine Learning, and Federated Explainable AI

Abstract

The prevalence of DDoS attacks has increased dramatically due to the proliferation of Internet of Things (IoT) devices, necessitating drastic detection methods that are beyond the capabilities of conventional deep learning models. The most recent advancements in DDoS detection techniques are outlined in this overview of the literature, with a focus on three novel paradigms: (1) Transformer-based architectures that use self-attention mechanisms to extract high-quality temporal-spatial features; (2) Quantum Machine Learning (QML) models that make use of quantum computational advantages to provide improved pattern recognition; and (3) Federated Learning with Explainable AI (XAI) that offers interpretable, privacy-preserving threat detection. We critically review 89 recent works (2023-2025), introducing new taxonomies, mathematical models, and algorithmic inventions that outperform the traditional LSTM-based ones. In our comparative studies, transformer models are found to have the highest accuracy of 99.79% and an enhancement of 0.10% relative to CNN-based systems, whereas quantum neural networks are found to have 99.87% detection rates with much fewer computational requirements. Moreover, federated XAI allows distributed learning in diverse settings with heterogeneous IoTs and retains model interpretability using SHAP values and attention. This review reports the important research gaps, presents hybrid structures between quantum-transformer- federated learning paradigms, and provides future research directions on next-generation IoT security. The mathematization of mathematical foundations, algorithmic designs, and performance benchmarks provide scholars and professionals with a complete blueprint to create solid, scalable and intelligent DDoS detection systems.