@Article{cmc.2023.040290,
AUTHOR = {Arif Hussain Magsi, Ali Ghulam, Saifullah Memon, Khalid Javeed, Musaed Alhussein, Imad Rida},
TITLE = {A Machine Learning-Based Attack Detection and Prevention System in Vehicular Named Data Networking},
JOURNAL = {Computers, Materials \& Continua},
VOLUME = {77},
YEAR = {2023},
NUMBER = {2},
PAGES = {1445--1465},
URL = {http://www.techscience.com/cmc/v77n2/54780},
ISSN = {1546-2226},
ABSTRACT = {Named Data Networking (NDN) is gaining a significant attention in Vehicular Ad-hoc Networks (VANET) due
to its in-network content caching, name-based routing, and mobility-supporting characteristics. Nevertheless,
existing NDN faces three significant challenges, including security, privacy, and routing. In particular, security
attacks, such as Content Poisoning Attacks (CPA), can jeopardize legitimate vehicles with malicious content. For
instance, attacker host vehicles can serve consumers with invalid information, which has dire consequences, including road accidents. In such a situation, trust in the content-providing vehicles brings a new challenge. On the other
hand, ensuring privacy and preventing unauthorized access in vehicular (VNDN) is another challenge. Moreover,
NDN’s pull-based content retrieval mechanism is inefficient for delivering emergency messages in VNDN. In this
connection, our contribution is threefold. Unlike existing rule-based reputation evaluation, we propose a Machine
Learning (ML)-based reputation evaluation mechanism that identifies CPA attackers and legitimate nodes. Based
on ML evaluation results, vehicles accept or discard served content. Secondly, we exploit a decentralized blockchain
system to ensure vehicles’ privacy by maintaining their information in a secure digital ledger. Finally, we improve
the default routing mechanism of VNDN from pull to a push-based content dissemination using Publish-Subscribe
(Pub-Sub) approach. We implemented and evaluated our ML-based classification model on a publicly accessible
BurST-Asutralian dataset for Misbehavior Detection (BurST-ADMA). We used five (05) hybrid ML classifiers,
including Logistic Regression, Decision Tree, K-Nearest Neighbors, Random Forest, and Gaussian Naive Bayes.
The qualitative results indicate that Random Forest has achieved the highest average accuracy rate of 100%. Our
proposed research offers the most accurate solution to detect CPA in VNDN for safe, secure, and reliable vehicle
communication.},
DOI = {10.32604/cmc.2023.040290}
}