Towards a Lightweight Detection System Leveraging Ranking Techniques with Wrapper Feature Selection Algorithm for Selective Forwarding Attacks in Low power and Lossy Networks of IoTs

The advent of the Internet of Things (IoT) marks a significant shift in networking paradigms, offering numerous advantages across various applications owing to advancements in embedded system technologies and IPv6 compression. This evolution enables the integration of IP functionalities within Low Power and Lossy Networks (LLNs). However, as LLNs become increasingly integral to IoT ecosystems, they become prime targets for cyber threats. While Machine Learning (ML) has demonstrated potential in identifying irregularities within the Routing Protocol for LLNs (RPL), it encounters issues, including the handling of unbalanced datasets, identifying critical features of attacks, long processing times, and substantial power usage. To address the preceding issues, this study introduces a lightweight detection system encompassing three stages: data preparation, a lightweight feature selection mechanism, and Random Forest-based selective forwarding attack detection. The proposed system is evaluated using a grid-based RPL dataset under three distinct scenarios—varying by the location of the attacking node concerning the LLN’s root node. The findings demonstrate that the system performs optimally when the attacker is positioned midway to the root node, achieving accuracy, precision, recall, and F-measure scores of 86.2%, 87%, 84.3%, and 85.6%, respectively. The outcomes suggest that this lightweight approach not only surpasses existing ML techniques in terms of execution speed and resource efficiency but also in the number of features required and overall detection performance, establishing its efficacy in combating Selective Forwarding (SF) attacks in LLNs.

Hali tarjima qilinmagan