SmartBuoy: A Machine Learning-Based Detection Method for Interest Flooding Attacks in VNDN

Due to the advantages of multi-source, multi-path, and in-network caching, Named Data Networking (NDN) can improve the efficiency of data exchange in Vehicular Ad-hoc NETworks (VANETs). As Vehicular Named Data Networking (VNDN) has become a new paradigm for connected vehicles, it also introduces a new security issue. Since VNDN follows the ‘request-response’ communication mode, malicious nodes can flood many interest packets to occupy resources, block the network, and damage intelligent transportation applications. Therefore, fast detection of Interest Flooding Attacks (IFA) is an urgent problem in VNDN. Some researchers have transplanted the IFA detection methods in NDN to VNDN, but there are still defects such as single detection category, low accuracy rate, and high overhead. Aiming at the above defects, an IFA detection method based on machine learning is proposed and named SmartBuoy. The contribution of our work lies in the following three points. Firstly, we present an IFA traffic generation method based on three simulation tools and construct a fine-grained dataset by building a dynamic directional interface model (DDIM). Secondly, by analyzing the traffic through directional interfaces, we propose 22 new features for IFA detection. Finally, we design a Two-Stage Two-Dimension feature selection algorithm (TSTD) to construct the optimal feature subset for a selected classifier with a specified number of features. We have verified the effectiveness of SmartBuoy through experiments. The results show that: (1) Compared with the rule-based methods, SmartBuoy can obtain a higher detection accuracy. (2) The new features designed according to DDIM can improve the performance of classifiers. (3) TSTD can help classifiers achieve higher accuracy than the other two classical algorithms with a specified number of features.