Unsupervised anomaly detection in internet of vehicles via spectral-residual preprocessing and adversarial variational autoencoding

Document Type

Article

Source of Publication

Computers and Electrical Engineering

Publication Date

9-1-2026

Abstract

The Internet of Vehicles (IoV) forms a large-scale cyber-physical system exchanging continuous data for safety, autonomy, and traffic efficiency, but faces significant cybersecurity risks. For example, exploiting IoV connectivity, advanced attacks such as spoofing, replay, and DoS often evade traditional Intrusion Detection Systems (IDSs). While Variational Autoencoder (VAE)-based unsupervised methods offer promise, they struggle with noisy, non-stationary data and poor generalization. In this paper, we present SPARTA, a Spectral-Preprocessing and Adversarial Representation Training Architecture for fully unsupervised IoV anomaly detection. Using Spectral Residual preprocessing, SPARTA converts multivariate signals into saliency-enhanced sequences processed by a PSO-optimized adversarial VAE. Binary PSO tunes architectural hyperparameters, while a dual-discriminator mechanism mitigates overfitting and enforces latent alignment. On average, performance evaluation results show that SPARTA improves detection accuracy by 2%, F1-score by 2.8%, and reduces false positives by 2.5% over existing methods. It delivers a 7% F1 improvement under domain shifts and over 6% gains in noisy settings, maintaining consistent 5–7% performance advantages across real-world IoV scenarios.

ISSN

0045-7906

Publisher

Elsevier BV

Volume

137

Disciplines

Computer Sciences

Keywords

Internet of vehicles, Particle swarm optimization, Spectral residual preprocessing, Unsupervised anomaly detection, Variational autoencoder

Scopus ID

105044311749

Indexed in Scopus

yes

Open Access

no

Link to Full Text

Share

COinS