Intrusion Detection Method for Internet of Vehicles Based on Parallel Analysis of Spatio-Temporal Features.

Area of Science:

• Cybersecurity
• Network Security
• Deep Learning

Background:

• Internet of Vehicles (IoV) faces increasing network security challenges.
• Existing deep learning intrusion detection methods struggle with comprehensive spatio-temporal feature extraction, leading to high false-positive rates.
• Limitations in current methods hinder effective intrusion detection in dynamic IoV environments.

Purpose of the Study:

• To propose a novel intrusion detection method for IoV utilizing parallel spatio-temporal feature analysis (PA-STF).
• To address the limitations of sequential feature extraction in existing deep learning models for IoV.
• To enhance the accuracy and reduce false positives in IoV network intrusion detection.

Main Methods:

• Feature selection based on IoV traffic correlation to build an optimal feature subset.
• Parallel extraction of spatio-temporal features using Temporal Convolutional Network (TCN) and Long Short-Term Memory (LSTM).
• Fusion of parallel features via a self-attention mechanism, followed by detection using a multilayer perceptron.

Main Results:

• The PA-STF method demonstrated superior performance on benchmark datasets (NSL-KDD and UNSW-NB15).
• Achieved a reduction in false-positive rates by 1.95% on NSL-KDD and 1.57% on UNSW-NB15.
• Showcased improvements in overall accuracy and F1 score for intrusion detection.

Conclusions:

• The PA-STF method effectively extracts and fuses spatio-temporal features for robust IoV intrusion detection.
• Parallel analysis significantly outperforms traditional serial methods in detecting network threats.
• The proposed approach offers a promising solution for enhancing IoV network security and reliability.