Countering DDoS Attacks in SIP Based VoIP Networks Using Recurrent Neural Networks

Area of Science:

• Computer Science
• Cybersecurity
• Network Engineering

Background:

• Voice over IP (VoIP) systems are increasingly adopted, but are vulnerable to security threats.
• Session Initiation Protocol (SIP), a core VoIP protocol, lacks inherent security features.
• Distributed Denial of Service (DDoS) attacks disrupt services by overwhelming VoIP systems.

Purpose of the Study:

• To develop an effective method for detecting DDoS attacks in VoIP networks.
• To enhance feature extraction from SIP messages for improved attack classification.
• To evaluate the performance of a deep learning model for DDoS detection.

Main Methods:

• Formulating DDoS detection as a classification problem.
• Utilizing token embedding to enhance features extracted from SIP messages.
• Implementing a deep learning model based on Recurrent Neural Networks (RNNs).
• Validating the system with a real traffic dataset including diverse attack scenarios.

Main Results:

• The proposed system achieves high detection accuracy for DDoS attacks.
• The system demonstrates a low detection time.
• Detection accuracy for low-rate attacks surpasses traditional machine learning methods.
• The RNN-based model effectively handles both low and high-rate DDoS attacks.

Conclusions:

• The token embedding and RNN-based approach offers a robust solution for VoIP security.
• This method provides a significant advancement in detecting DDoS attacks, particularly low-rate variants.
• The findings contribute to securing VoIP communications against evolving cyber threats.