Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

FAST-IDS: A Fast Two-Stage Intrusion Detection System with Hybrid Compression for Real-Time Threat Detection in Connected and Autonomous Vehicles

Devika S, Vishnu Hari, Pratik Narang, Tejasvi Alladi, Vinay Chamola

TL;DR

This work introduces FAST-IDS, a two-stage intrusion detection framework tailored for resource-constrained Connected and Autonomous Vehicles (CAVs). Stage 1 employs a Bidirectional GAN with Wasserstein loss and gradient penalty for coarse anomaly detection, while Stage 2 uses a hybrid CNN-LSTM for fine-grained multiclass classification across 19 attack types, all under a hybrid model compression pipeline (structured pruning + static quantization). The approach achieves strong unseen-attack detection, and substantial reductions in model size and inference time (≈77.2% size reduction and ≈50.05% faster inference), enabling per-vehicle threat detection on Jetson Nano (~0.195 s). Experimental results on the VeReMi Extension dataset demonstrate competitive Stage 2 multiclass accuracy (≈97.9%), Stage 1 recall improvements over SOTA, and robust performance across both known and unseen attacks, highlighting its practical impact for real-time ITS security.

Abstract

We have implemented a multi-stage IDS for CAVs that can be deployed to resourec-constrained environments after hybrid model compression.

FAST-IDS: A Fast Two-Stage Intrusion Detection System with Hybrid Compression for Real-Time Threat Detection in Connected and Autonomous Vehicles

TL;DR

This work introduces FAST-IDS, a two-stage intrusion detection framework tailored for resource-constrained Connected and Autonomous Vehicles (CAVs). Stage 1 employs a Bidirectional GAN with Wasserstein loss and gradient penalty for coarse anomaly detection, while Stage 2 uses a hybrid CNN-LSTM for fine-grained multiclass classification across 19 attack types, all under a hybrid model compression pipeline (structured pruning + static quantization). The approach achieves strong unseen-attack detection, and substantial reductions in model size and inference time (≈77.2% size reduction and ≈50.05% faster inference), enabling per-vehicle threat detection on Jetson Nano (~0.195 s). Experimental results on the VeReMi Extension dataset demonstrate competitive Stage 2 multiclass accuracy (≈97.9%), Stage 1 recall improvements over SOTA, and robust performance across both known and unseen attacks, highlighting its practical impact for real-time ITS security.

Abstract

We have implemented a multi-stage IDS for CAVs that can be deployed to resourec-constrained environments after hybrid model compression.
Paper Structure (33 sections, 17 equations, 6 figures, 8 tables, 2 algorithms)

This paper contains 33 sections, 17 equations, 6 figures, 8 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Intrusion Detection Systems in CAVs
  4. Computational Efficiency in Resource Constrained Environments
  5. GANs for Intrusion Detection
  6. Detection of Unseen Attacks
  7. Multistage Intrusion Detection
  8. Preliminary Background
  9. CAV Scenario
  10. Dataset
  11. Taxonomy of Attacks
  12. Proposed Intrusion Detection Framework
  13. Stage 1
  14. Data Preprocessing
  15. Architectures Employed
  16. ...and 18 more sections

Figures (6)

  • Figure 1: A typical CAV scenario.
  • Figure 2: Proposed FAST-IDS architecture.
  • Figure 3: Whisker boxplots of reconstruction error for the seven BiGAN model variants used in the ablation study in Stage 1. Each boxplot illustrates the IQR ranges between 1st quartile (Q1) and the 3rd quartile (Q3) and extends upto the Lower (LL) and Upper Limits (UL).
  • Figure 4: Graph illustrating the performance of FAST-IDS (Stage 1) with respect to other SOTA baselines in terms of recall percentages.
  • Figure 5: The confusion matrix visualized as the heat map of the Stage 2 Hybrid CNN-LSTM model evaluated across 19 attack classes.
  • ...and 1 more figures