Towards a Machine Learning Driven Trust Management Heuristic for the Internet of Vehicles

Area of Science:

• Cybersecurity
• Internet-of-Vehicles (IoV)
• Machine Learning

Background:

• Conventional trust assessment schemes in IoV struggle with insider attackers.
• Accurate weighting of trust attributes and defining trust thresholds are critical for reliable detection.
• Existing methods often rely on aggregating individual trust attributes.

Purpose of the Study:

• To develop and evaluate an improved trust assessment scheme for the Internet-of-Vehicles (IoV) environment.
• To compare the effectiveness of different feature matrix formations in classifying honest and dishonest vehicles.
• To identify optimal machine learning algorithms for IoV trust assessment.

Main Methods:

• Transformed an IoT dataset (CRAWDAD) into an IoV format, including 18,226 interactions among 76 nodes.
• Computed influencing parameters: packet delivery ratio, familiarity, timeliness, and interaction frequency.
• Created two feature matrices (FM1 and FM2) and applied unsupervised and supervised machine learning (Subspace KNN, Subspace Discriminant) for classification.

Main Results:

• Subspace KNN achieved perfect precision, recall, and F1-score (1) for individual parametric scores.
• Subspace Discriminant achieved perfect precision, recall, and F1-score (1) for mean parametric scores.
• Feature matrix FM2 demonstrated superior classification accuracy compared to FM1.

Conclusions:

• The proposed approach using feature matrix FM2 and machine learning effectively enhances trust assessment in IoV.
• Averaging pairwise computations (FM2) provides a more robust feature set for detecting malicious nodes.
• The study provides valuable insights into optimizing trust thresholds and feature engineering for IoV security.