Attack types and their description.

Vehicular Ad-hoc Networks (VANETs) are critical to Intelligent Transportation Systems (ITS), enabling vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) communication to improve road safety and traffic flow. However, VANETs face significant security threats, particularly position falsification attacks, where malicious nodes disseminate false Basic Safety Messages (BSMs). This study proposes an ensemble learning framework to detect such attacks, leveraging Decision Tree (CART), Random Forest, K-Nearest Neighbors (KNN), and Multilayer Perceptron (MLP) classifiers enhanced with bagging. Using the VeReMi dataset, our RSU-level detection system analyzes sequential BSMs to detect malicious behavior. Results demonstrate that KNN with bagging achieves perfect precision, recall, accuracy, and F1 score (100%) for Attack 1, while maintaining near-perfect performance for complex attacks like Attack 2 (99.87% accuracy) and Attack 16 (97.85% accuracy). Decision Tree with bagging also performs well for simpler attacks but experiences a slight decline for highly complex scenarios. Random Forest with bagging excels in simpler attacks but struggles with complex patterns. MLP with bagging shows strong results for simpler attacks but underperforms in complex scenarios. The proposed framework highlights the effectiveness of ensemble techniques, particularly KNN with bagging, in safeguarding VANET communication systems, offering a scalable, efficient, and robust solution for VANET security.

车载自组织网络（Vehicular Ad-hoc Networks, VANETs）对智能交通系统（Intelligent Transportation Systems, ITS）至关重要，其支持车车通信（Vehicle-to-Vehicle, V2V）与车路通信（Vehicle-to-Infrastructure, V2I），旨在提升道路安全水平与交通运行效率。然而，车载自组织网络面临诸多严峻的安全威胁，尤以位置伪造攻击为甚——此类攻击中，恶意节点会传播虚假的基本安全消息（Basic Safety Messages, BSMs）。本研究提出一种集成学习框架以检测此类攻击，该框架融合了经装袋法（bagging）增强的决策树（Decision Tree, CART）、随机森林、k近邻算法（K-Nearest Neighbors, KNN）以及多层感知器（Multilayer Perceptron, MLP）分类器。本研究依托VeReMi数据集，构建了路侧单元（Road Side Unit, RSU）级别的检测系统，通过分析连续的基本安全消息以识别恶意行为。实验结果表明，搭载装袋法的k近邻算法在攻击1上实现了完美的精确率、召回率、准确率与F1值（均为100%），且在攻击2（准确率99.87%）与攻击16（准确率97.85%）等复杂攻击场景下仍保持近乎完美的性能。搭载装袋法的决策树在简单攻击场景中表现优异，但在高度复杂的场景中性能略有下降；搭载装袋法的随机森林擅长处理简单攻击，但难以应对复杂模式；搭载装袋法的多层感知器在简单攻击场景中效果出色，但在复杂场景中性能欠佳。本研究提出的框架彰显了集成学习技术的有效性，尤其是搭载装袋法的k近邻算法，在保障车载自组织网络通信系统安全方面表现突出，可为车载自组织网络安全提供一种可扩展、高效且鲁棒的解决方案。