Enhancing convolutional neural network - based distributed denial of service attack (DDoS) Detection systems using L2 and dropout regularization in web servers. 6

Undergraduate Thesis : (Bachelor of Science major in Computer Science) - Pamantasan ng Lungsod ng Maynila, 2024. 56

5

ABSTRACT: Distributed Denial of Service (DDoS) attacks are pervaqsive and destructive cyber threats that distrupt online services and infrastructures. Malicious actors coordinate a barrage of requests from multiple compromised devices, overwhelming tarfets and rendering them inaccessible. The impacts are profound, causing extended downtime and significant financial losses for businesses. Additionally, DDoS attacks tarnish reputations, eroding trust in affected entities services. While convolutional Neural Networks (CNN) have been used as a deterrent to such attacks, they alone may not be sufficient to combat the envolving sophistication of DDoS threats. These attacks remain a persistent threat due to challenge such as high false positive rates, compromises in model size and complexity, and the lack of frequent model updates. Effective mitigation strategies are crucial to safeguarding against these pervasive cyber threats and ensuring the resilience pf online infrastructures. In response, our study aims to enhance CNN-based DDoS detection systems in web servers by integratin g L2 and Dropout regularization techniques and implementing a batch model updating process. Leveraging packet capture data, we train and evaluate our enhanced detection system to address these challenges. Our results demonstrate a significant improvement in detection accuracy, with the enhanced modelachieving an accuracy of 9.917 compared to the existing system's 0.863. Future research explores the integration of anomaly detection techniques alongside the CNN-based detection system thereby strengthening the defense mechanisms against evolving cyber threats.

5