HiMSELF: A Hierarchical Misbehavior Classification With Sequence Embedding by Latent Features in Vehicular Ad-Hoc Networks
Mingyu Kim, Dae Hyun Yum, Jaehee Jung · IEEE Access · 2025
Vehicular ad hoc network (VANET) enables vehicles, infrastructure, and pedestrians to exchange information, enhancing safety, efficiency, and intelligent decision-making. VANETs operate by exchanging basic safety messages (BSMs) between vehicles and infrastructure. However, as network connectivity increases, so does the risk of malfunctions or malicious data injections that can cause severe disruptions. Accordingly, a misbehavior detection system (MDS) to detect and analyze abnormal data patterns and potential attack behaviors has become essential. In particular, identifying misbehavior classification based on specific abnormal types is essential for establishing a robust and comprehensive VANET security framework. To address these challenges, a hierarchical misbehavior classification method with sequence embedding by latent features, referred to as HiMSELF, is proposed. HiMSELF, defined as a hierarchical classification system (HCS), employs a trained deep learning (DL) model to classify various misbehavior types, embed their intrinsic representations, and perform hierarchical clustering. The resulting structure reflects the intrinsic relationships among the types and serves as the basis for constructing the HCS. The HiMSELF classification pipeline operates in two sequential stages. First, the system employsBSMsequence data as input and grouped into broader, higher-level categories associated with each sample. Subsequently, it identifies the specific misbehavior type within the corresponding category. In experiments classifying 19 misbehavior types, HiMSELF achieved an average F1-score of 0.9918, outperforming existing approaches and demonstrating its potential to underpin reliable security mechanisms in cooperative intelligent transportation systems (C-ITS).