A Machine Learning-based Scheme for Enhancing the Detection of Position Falsification Attacks in Vehicular Ad-Hoc Networks

Abdelkreem, Eslam; Hussein, Sherif; Tammam, Ashraf · 2024 · Crossref

DOI: 10.21203/rs.3.rs-4516995/v2

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the security vulnerability of Vehicular Ad-Hoc Networks (VANETs) to position falsification attacks, where malicious nodes broadcast spoofed location data to disrupt safety applications and potentially cause accidents. Existing Misbehavior Detection Schemes (MDSs) often rely on application-layer features that attackers can manipulate to evade detection. To overcome this limitation, the authors propose a machine learning-based detection scheme that incorporates a novel physical-layer feature, termed RSSIConf, which utilizes the Received Signal Strength Indicator (RSSI) and its confidence intervals at varying distances to verify the plausibility of claimed locations. The study employs the Vehicular Reference Misbehaviour Dataset (VeReMi), which simulates five types of position falsification attacks. The methodology involves preprocessing the dataset and deriving the RSSIConf feature by calculating RSSI confidence boundaries for specific distance ranges at 90%, 95%, and 99% confidence levels. This new feature is combined with standard features (e.g., sender location, speed, and variance metrics) to create three feature vectors. The authors train and evaluate three machine learning algorithms—Random Forest (RF), K-Nearest Neighbors (KNN), and Multi-Layer Perceptron (MLP)—using Binary Classification Mode (BCM) to distinguish between benign and malicious messages. Performance is assessed using accuracy, precision, recall, and F1-score. The results indicate that the proposed model significantly outperforms existing approaches. Specifically, the inclusion of the RSSIConf feature improved detection accuracy by 0.76% to 13.26% and F1-Score by 0.74% to 12.71%, depending on the attack type. Among the tested algorithms, Random Forest consistently achieved the highest performance, reaching near-perfect metrics (e.g., 99.99% accuracy) for certain attack types like Type-1 (fixed position) and Type-2 (constant offset). The study demonstrates that integrating physical-layer signal characteristics with machine learning provides a more robust defense against position spoofing than relying solely on message content analysis. The significance of this work lies in its contribution to VANET security by introducing a feature that is difficult for attackers to falsify, as it depends on the physical propagation properties of the wireless channel. By validating the model against state-of-the-art benchmarks using a standardized dataset, the authors provide a reproducible framework for enhancing misbehavior detection. The findings suggest that hybrid feature sets combining application and physical layers are essential for developing resilient security mechanisms in intelligent transportation systems.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-25; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• v2x connected vehicle