مهند البوراضي

Software-Defined Networking (SDN) changes how networks are managed by putting all control in one place. However, this setup has a major flaw: the controller is a single point of failure for Distributed Denial-of-Service (DDoS) assaults. Traditional signature-based methods donʹt work against zero-day threats, an‎d supervised machine learning methods often need too much computing resources an‎d large labeled datasets that arenʹt available for new assaults. This thesis provides a Lightweight Semi-Supervised Hybrid PCA-K-Means framework designed to efficiently detect unexpected DDoS attacks within the resource-constrained SDN control plane. The method uses Principal Component Analysis (PCA) to cut the number of features by about 54% while keeping 99% of the data variance. It also uses a new "seeded" K-Means clustering algorithm. This method uses a stratified subset (10%) of labeled data to smartly set the initial cluster centroids, which makes sure that the results are semantically correct without the need for full supervision. The modelʹs robustness was tested on the CICDDoS2019 an‎d InSDN datasets, where it was able to find 82.75% an‎d 87.88% of the correct answers, respectively, with a false positive rate as low as 0.59%. The system is very efficient at processing data, han‎dling more than 8 million flows per second with latency of less than a millisecond on conventional CPU hardware. These results show that the suggested hybrid model effectively combines the accuracy of supervised learning with the flexibility of unsupervised methods. This makes it a useful, real-time solution for dealing with zero-day threats in high-speed networks.

يادگيري نيمه‌نظارتي , DDoS، ، خوشه‌بندي , شبكه نرم‌افزارمحور (SDN)، , تشخيص نفوذ سبك

: Software-Defined Networking (SDN , : Software-Defined Networking (SDN , DDoS Detection, , Semi-Supervised Learning , K-Means Clustering

Mohanad alboradhe

Dr. Javad Vahidi