P3IDF-EC: PCA-Based Privacy-Preserving Intrusion Detection Framework for Edge Computing

In the modern era, the use of edge computing devices is increasing and being used rapidly. This trend leads to the increasing presence of sensitive/personal data in edge environments. This is mainly due to the key functionalities, such as real-time processing, storing data on edge networks, reducing latency, and using higher bandwidth of edge computing. The main goal of edge-based computing is to continuously offer users the highest possible quality of service and experience, achieved through the mobility of edge devices. The mobility of edge devices presents significant challenges, particularly in terms of user data privacy and data leakage risks. Additionally, the resource-constrained nature results in edge devices being interconnected into less secure networks, and as well as the decentralized nature expands the potential target for an attack, which lures in both malicious and inquisitive intruders. As such systems are increasingly deployed with limited resources, there is a need to develop robust privacy-preserving intrusion detection techniques to safeguard original data and identify network cyber threats. To maintain a good balance between security and privacy, it is imperative to protect the privacy of user data in edge computing environments. In this light, a principal component analysis-based privacy-preserving intrusion detection framework for edge computing, so-called P3IDF-EC is proposed to achieve robust intrusion detection while providing privacy preservation for sensitive data through machine learning approaches. The proposed model is trained on the benchmark UNSW-NB 15 dataset, and experimental results highlight that the framework efficiently secures user data on edge-based networks, and performs accurate intrusion behavior determination. Compared with the current privacy-based cutting-edge techniques, P3IDF-EC outperforms them all by achieving an accuracy (ACC) of 99.84%, and a false positive rate (FPR) of 0.13%. Also exhibits a high privacy index of 97%, and less computational processing time (CPT).