ML-based categorical boosting with hybrid transfer learning model for enhancing cyber threat intelligence in IoV environment

The increasing complexity of the Internet of Vehicles (IoV) necessitates robust Intrusion Detection Systems (IDS) to protect against cyberattacks. Recent studies show a 60% rise in attacks targeting vehicular communication systems, with 45% attributed to attacks. Existing IoV-based IDS methods face...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Journal of Umm Al-Qura University for Engineering and Architecture
Hauptverfasser: Supriya, Dhanda, Praveen Krishna, Anne Venkata
Format: Journal Article
Sprache:Englisch
Veröffentlicht: 17.10.2025
ISSN:2731-6688, 1658-8150
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:The increasing complexity of the Internet of Vehicles (IoV) necessitates robust Intrusion Detection Systems (IDS) to protect against cyberattacks. Recent studies show a 60% rise in attacks targeting vehicular communication systems, with 45% attributed to attacks. Existing IoV-based IDS methods face challenges such as imbalanced datasets and limited feature extraction capabilities, which hinder accurate threat detection. The proposed IoV-Net framework addresses these issues by integrating comprehensive data preprocessing, where noise and redundancies in the Canadian Institute for Cybersecurity collected IoV 2024 (CICIoV2024) dataset are eliminated, followed by dimensionality reduction. For feature extraction, the Transfer Learning Adopted Hybrid Inception-ResNetV2 (TLA-HIR) model is employed, leveraging its superior ability to capture both local and global patterns in IoV data. To tackle class imbalance, the Adaptive Synthetic Minority Over-Sampling (ASMOS) technique is introduced, enhancing the dataset with synthetic samples for underrepresented classes, thus preventing overfitting. Finally, the Machine Learning-based Categorical Boosting (MLCB) Classifier is implemented, ensuring high accuracy in attack classification by utilizing gradient boosting and efficient handling of categorical features. This methodology, applied to the CICIoV2024 dataset with the feature analysis on three cases: binary, decimal, and hexadecimal, promises improved detection rates for sophisticated IoV-based attacks with an accuracy of 99.84%, 99.88%, and 99.88%, respectively.
ISSN:2731-6688
1658-8150
DOI:10.1007/s43995-025-00225-x