Enhancing DDoS Attack Classification through SDN and Machine Learning: A Feature Ranking Analysis

Due to the growing dependence of digital services on the Internet, Distributed Denial of Service (DDoS) attacks are a common threat that can cause significant disruptions to online operations and financial losses. Machine learning (ML) offers a promising way for early DDoS attack detection due to it...

Full description

Saved in:
Bibliographic Details
Main Authors: Aymen AlAwadi, Kawthar Rasoul ALesawi
Format: Article
Language:English
Published: Faculty of Engineering, University of Kufa 2025-04-01
Series:Mağallaẗ Al-kūfaẗ Al-handasiyyaẗ
Subjects:
Online Access:https://journal.uokufa.edu.iq/index.php/kje/article/view/16742
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Due to the growing dependence of digital services on the Internet, Distributed Denial of Service (DDoS) attacks are a common threat that can cause significant disruptions to online operations and financial losses. Machine learning (ML) offers a promising way for early DDoS attack detection due to its ability to analyze large datasets and identify patterns. However, adding too many features to the ML might reduce its effectiveness in identifying the attacks provided by central network paradigms such as the Software-Defined Network (SDN). In this research, we investigate the effectiveness of the ML methods such as (Random Forest (RF), Naive Base (NB), and K-Nearest Neighbor’s (KNN)) combining SDN to enhance the classification of DDoS attacks. We leverage three diverse datasets: DDoS attack SDN, CICDDoS2019, and SDN-DDOS-TCP-SYN dataset. By leveraging cross-feature selection and feature ranking techniques, such as information gain, gain ratio, and Gini importance, we could identify the most relevant network features for DDoS attacks. We reduced the feature up to 5 effective features without compromising the classification accuracy. The experimental results show that the proposed models achieved an accuracy of 100% for both Random Forest (RF) and K-Nearest Neighbor (KNN), and 99.8% for Naive Bayes (NB). Due to their high accuracy and lower complexity, KNN and NB outperform ML algorithms in this study
ISSN:2071-5528
2523-0018