Implementing learning-based (ML-based) hybrid model to mitigate distributed denial of service (DDoS) attacks in mobile edge computing (MEC)

Abstract

5G technology constitutes a considerable part of solving the problem of security in mobile communications. Multi-Access Edge Computing or mobile edge computing (MEC) extends the capabilities of cloud computing by locating them near the edge of the network. By outsourcing cloud processing to specific local servers, MEC decreases latency in 5G, thereby improving the end-user experience. This study explores a security vulnerability present in 5G MEC. Specifically, we examined distributed denial of service (DDoS) attacks occurring at both the network and the application layer. The vulnerability of MEC to DDoS attacks poses significant challenges that are addressed in this research. We evaluated different Machine Learning (ML) algorithms and subsequently implemented hybrid models (Stacking/Blending, and Random Forests (RF) model) which are classified under supervised ML. The purpose of this study is to identify the most effective techniques for mitigating DDoS attacks in MEC systems. ML techniques such as Random Forest (RF), Decision tree (DT), Naïve Bayes (NB), K Nearest Neighbour (K-NN), Logistics regression (LR), Blending/Stack Model are evaluated on the basis of a variety of performance metrics (including accuracy, detection/recall, precision, F1-Measure, Matthews correlation coefficient (MCC), Receiver operating characteristic (ROC), and Area Under Receiver operating characteristic (AUROC)) for each of the algorithms. Probability density function (PDF) and hypotheses testing are statistical techniques deployed to support the findings of our study. Based on the literature in the field, ML techniques are recommended to reach our solution. The best ML algorithms yielding the best performance in mitigating the DDoS attacks are optimized to enhance their performance ability. This study outlines the overview of MEC environment’s existing mitigation scheme, and the implemented mitigation schemes towards DDoS attacks. According to our evaluated findings, Hybrid models outperformed ML models based on the computed scores of performance metrics. PDF and hypotheses testing successfully supported our findings by showing that hybrid models indeed outperformed ML models. Among the mitigation techniques, RF outperformed all supervised ML models by effectively mitigating DDoS attacks in MEC.