A Dynamic Hybrid Stacking Approach for Real-Time Threat Detection in High-Mobility Vehicular Networks

Abstract

Vehicular Ad Hoc Networks (VANETs) have become a fundamental part of modern intelligent transportation systems, enabling vehicles to communicate for improved safety and traffic management. Historically, early network security mechanisms in vehicular systems relied on manual monitoring and rule-based approaches to identify malicious activities. However, with the rapid expansion of connected vehicles, these traditional manual systems are no longer sufficient to handle complex and large-scale network data. The major problem lies in accurately detecting malicious nodes involved in attacks such as Denial of Service (DoS), Sybil, and Blackhole, which can severely disrupt communication and compromise system reliability. Manual detection methods are time-consuming, error-prone, and lack scalability, making them ineffective in dynamic and high-mobility VANET environments. This creates a strong need for automated, intelligent, and adaptive detection systems. To address this challenge, the proposed system introduces a Flask-based malicious node detection framework that leverages machine learning models including AdaBoost (AB), Logit Boosting (LB), Gradient Boosting (GB), and Natural Gradient Boosting (NGB). The system incorporates automated preprocessing, real-time exploratory data analysis, and multi-target classification to improve detection accuracy. Among these, NGB demonstrates superior performance, achieving accuracies of 97.61% for DoS attacks, 97.96% for Sybil attacks, and 96.70% for Blackhole attacks due to its probabilistic learning capability and ability to handle uncertainty. The proposed system significantly enhances detection efficiency and accuracy, providing a scalable and user-friendly solution that strengthens network security and ensures reliable communication in intelligent vehicular environments.