A novel optimization-driven deep learning framework for the detection of DDoS attacks.

Area of Science:

• Computer Science
• Cybersecurity
• Artificial Intelligence

Background:

• Distributed Denial of Service (DDoS) attacks pose a significant threat to cloud computing and network services.
• These attacks lead to service unavailability, causing substantial financial and reputational damage.
• Effective threat identification is crucial for mitigating revenue loss and enhancing brand reputation.

Purpose of the Study:

• To implement an effective intrusion detection system (IDS) for identifying DDoS attacks.
• To leverage deep learning techniques for enhanced threat detection in cloud environments.
• To minimize bias in datasets and improve the accuracy of attack classification.

Main Methods:

• A three-phase framework: Data pre-processing, Data balancing using Conditional Generative Adversarial Networks (CGAN), and Classification.
• Classification is performed using a Stacked Sparse Denoising Autoencoder (SSDAE) optimized by a Firefly-Black Widow (FA-BW) hybrid algorithm.
• Experiments were validated using the CICDDoS2019 dataset and compared against existing techniques.

Main Results:

• The proposed deep learning framework demonstrated significantly higher accuracy in detecting DDoS attacks compared to other methods.
• Conditional GANs effectively minimized bias towards majority classes in the dataset.
• The SSDAE with FA-BW optimization proved effective in distinguishing between attack and benign traffic.

Conclusions:

• Advanced deep learning techniques and hybrid optimization algorithms play a vital role in strengthening cybersecurity against DDoS threats.
• The developed intrusion detection system offers a robust solution for mitigating the impact of DDoS attacks.
• This research underscores the potential of AI-driven approaches in proactive network security.