
Temporal Hybrid Neural Network Achieves Over 99 Percent DDoS Detection Accuracy - Bioengineer.org
Researchers Mohammad Shahid and Mukesh Mann have developed a neural network that detects distributed denial-of-service attacks with over 99 percent accuracy.
Software-defined networking separates administrative routing from physical switches, creating centralized control points that are vulnerable to flooding attacks. Malicious traffic can overwhelm flow tables, degrading service for legitimate users.
The researchers integrated the Synthetic Minority Oversampling Technique to correct dataset imbalances and a feature selection routine named Hybrid Crayfish-Gazelle Optimization. They combined these with a Temporal Hybrid Attention Recurrent Network to create an end-to-end framework.
The system was evaluated offline and achieved an overall classification accuracy of 99.38 percent, exceeding conventional deep learning baselines. This performance highlights the potential of advanced machine learning to secure modern cloud infrastructure.
Addressing Limitations
Conventional defenses like static firewalls struggle to adapt to evolving traffic structures in programmable environments. Many existing intrusion detection models rely on static flow metrics and overlook severe class imbalances.
The new model addresses these issues by modeling temporal traffic fluctuations and reducing feature redundancies. This approach helps mitigate biased predictions and high false positive rates in resource-limited controllers.