A Survey of Semi-Supervised Learning Auto-Encoders and Probabilistic Bayesian Network Approaches
Cybersecurity threats have evolved in complexity and scale, demanding adaptive and intelligent detection mechanisms. This survey examines the integration of Semi-Supervised Learning (SSL) Auto-Encoders and Probabilistic Bayesian Networks (PBNs) as hybrid models for effective cyber risk management and attack detection. This synthesizes recent scholarly contributions (2020–2025) exploring machine learning, deep learning, and probabilistic reasoning approaches. Results indicate that SSL-AE and PBN integration enhances detection accuracy, scalability, and interpretability, addressing the data scarcity problem inherent in supervised models. A comparative analysis of recent models reveals that hybrid frameworks achieve average precision rates exceeding 93% on benchmark datasets, such as UNSW-NB15. This review identifies persistent challenges in model explainability, real-time deployment, and adversarial robustness, while projecting future trends in self-learning cyber defense systems that leverage generative AI and Bayesian deep fusion models.
Download Paper