A Multi-Modal Deep Transfer Learning Framework for Attack Detection in Software-Defined Networks

Abstract

Software-defined networking (SDN) has been recognized for its potential in network programming and centralized control. However, this advancement brings forth critical security vulnerabilities. It is essential to understand that vulnerabilities, by their inherent nature, may lead to potential attacks if not addressed timely and appropriately. In this paper, we introduce a novel multi-modal deep transfer learning (MMDTL) framework tailored for effective attack detection in SDN environments that helps us to investigate a diverse spectrum of attack types. MMDTL framework comprehensively incorporates diverse data modalities - encompassing network traffic patterns, system logs, and user behavior analytic. A pivotal feature of this framework is its transfer learning approach, which enables the assimilation of insights from pre-trained models that subsequently increases the detection performance of attacks. We rigorously analyze the proposed framework with experiments on the intrusion detection evaluation dataset (CIC-IDS2017). Analyses results show the superiority of our framework with a detection accuracy 99.97%.Thus, MMDTL framework has a significant potential to support security in SDNs.

Keywords

• Attack detection
• CICIDS2017
• data analysis
• transfer learning
• network programming
• software-defined network