Adaptive DDoS detection mode in software-defined SIP-VoIP using transfer learning with boosted meta-learner

Area of Science:

• Cybersecurity
• Machine Learning
• Network Security

Background:

• The internet's infrastructure is vulnerable to sophisticated cyberattacks like distributed denial-of-service (DDoS).
• Existing security measures like firewalls offer limited protection against evolving threats.
• Traditional machine learning (ML) methods struggle with categorical data and local maxima, while deep learning (DL) faces challenges like vanishing gradients and long training times.

Purpose of the Study:

• To develop an advanced, efficient method for detecting and mitigating DDoS attacks.
• To overcome the limitations of traditional ML and DL techniques in network security.

Main Methods:

• Proposed a novel transfer learning scheme employing a 3-base classifier ensemble (BiGRU, BiLSTM, Random Forest) integrated with an XGBoost meta-learner.
• Utilized this ensemble to analyze and classify network traffic for identifying malicious packets indicative of DDoS attacks.

Main Results:

• The proposed ensemble achieved perfect Accuracy and F1 scores of 1.000.
• Successfully classified 314,102 DDoS cases during evaluation, demonstrating high efficacy.
• The system efficiently identified malicious packets associated with DDoS attacks in network transactions.

Conclusions:

• The developed transfer learning ensemble effectively identifies malicious packets for DDoS attacks.
• This advanced machine learning approach offers a robust solution for enhancing network security against sophisticated threats.