Enhancing security in software defined networks: Privacy-preserving intrusion detection with Homomorphic Encryption

Abstract

This study proposes a novel privacy-preserving intrusion detection framework for software-defined networks (SDNs) by integrating Homomorphic Encryption (HE) with Deep Neural Networks (DNNs). The framework encrypts network traffic using HE before performing intrusion detection analysis with a DNN model, ensuring data confidentiality while enabling robust threat detection. The proposed approach involves encrypting the dataset, training the DNN-based intrusion detection model on encrypted data, and deploying the model within the SDN architecture. Key findings demonstrate that the DNN achieves high accuracy (87.11 %) on encrypted data, comparable to its performance on unencrypted data (99.99 %), indicating its suitability for secure applications. In contrast, traditional machine learning models such as Logistic Regression, Random Forest, and Decision Tree exhibit decreased accuracy on encrypted data compared to their performance on unencrypted data. The minimal performance difference of the DNN between encrypted and unencrypted datasets highlights its effectiveness for applications prioritizing security and privacy. The proposed framework incorporates encryption at critical stages, from data collection to application deployment, and leverages robust control mechanisms like SDN controllers and open flow switches to strengthen the overall security posture. This study represents a significant step towards achieving privacy-preserving intrusion detection in SDNs, contributing to ongoing efforts to enhance network security while safeguarding data privacy against evolving cybersecurity threats.