Deep Learning-Driven Real-Time Anomaly Detection in SDNs: a Performance Comparison

EasyChair Preprint 15583

Abstract

In this paper, we explore and advance deep learning algorithms for anomaly detection in Software Defined Networks (SDN). As SDNs gain prominence in modern networking, their centralized and dynamic nature exposes them to threats like DDoS attacks and unauthorized access. Traditional detection methods often struggle to address these challenges, prompting the need for adaptive solutions. This study evaluates Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and Autoencoders for their effectiveness in detecting anomalies in SDN environments. Through extensive experiments, we compare their performance metrics, highlighting CNNs’ strength in spatial anomalies, RNNs’ suitability for temporal patterns, and Autoencoders’ ability to detect unseen anomalies. We also assess threshold sensitivity and real-time feasibility. Our findings demonstrate that deep learning significantly enhances SDN security, providing accurate and fast anomaly detection. Finally, we propose future directions for scaling these models to dynamic, large-scale SDN deployments.

Keyphrases: Convolutional Neural Networks (CNN), Network Security, Software Defined Networking (SDN), anomaly detection, deep learning, machine learning

@booklet{EasyChair:15583,
  author    = {Hoo Wang and Che Leo and John Davis and Sarah Smith and Daniel Taylor and Michael Lornwood},
  title     = {Deep Learning-Driven Real-Time Anomaly Detection in SDNs: a Performance Comparison},
  howpublished = {EasyChair Preprint 15583},
  year      = {EasyChair, 2024}}