Intrusion Detection Systems Based on Stacking Ensemble Learning in VANET

Abstract

Vehicular ad-hoc network (VANET) will play an important role in improving driving safety and efficiency in transport systems. As various attacks arise in VANET, it is essential to design mechanisms that can detect these attacks and then mitigate them. In this paper, we make an effort to detect five different position falsification attacks in VANET, including constant attack, constant offset attack, random attack, random offset attack, and eventual stop attack. Two detection systems based on ensemble machine learning algorithms, including stacking ensemble learning algorithms for classification and stacking ensemble learning for neural network, are proposed. We extracted the most important features by performing feature importance techniques. Then, we train the proposed learning algorithms on VeReMi dataset which includes five different position falsification attacks with three traffic densities and three attacker densities. Extensive experimental results are provided to evaluate the proposed solutions' effectiveness. Based on our results, stacking ensemble learning for classification algorithm can achieve the best performance in terms of accuracy and recall. In low density traffic, accuracy and recall of stacking ensemble learning for classification algorithms are 1 for the constant attack, constant offset attack, and random attack. Accuracy and recall for the random offset attack are 0.999 and 0.996, respectively. For the eventual stop attack, accuracy and recall are 0.995 and 0.985, respectively. In medium density, accuracy and recall of stacking ensemble learning also achieve the best performance.