Mobility Discloses Genuinity: A Robust Machine Learning-Based Sybil Attack Detection Scheme

In this paper, a robust machine learning (ML)-based scheme for Sybil attack detection in mobile networks is proposed. The proposed scheme exploits three physical-layer features, namely the Doppler shift, received signal strength (RSS), and channel state information (CSI) for identifying Sybil nodes. By employing a Bayesian optimization method, an optimized Random Forest ML classifier is utilized for the classification phase by exploiting the estimated and processed physical-layer attributes, yielding an efficient node classification and Sybil attack detection in a mobile network. A thorough performance evaluation of the proposed scheme is performed in terms of its receiver operating characteristic (ROC) curve, demonstrating higher node classification accuracy gains. Furthermore, the proposed scheme outperforms its benchmark schemes, namely the single- and dual-attribute schemes and the three-features hypothesis-based one. Specifically, the proposed scheme improves the true positive rate (TPR) by 12.5% compared to its dual-feature RSS-Doppler shift-based counterpart, and enhances the Doppler shift-, RSS-, and CSI-based single-attribute ones by 216%, 137%, and 36%, respectively, in terms of the TPR.