Lightweight Deep Learning Intrusion Detection for Safer Connected Vehicles

What is it about?

Modern vehicles are increasingly connected, with many parts like brakes and engines communicating via a common network called the Controller Area Network (CAN). Unfortunately, this CAN bus has security weaknesses that can be exploited by hackers to launch stealthy attacks, which could compromise the safety of vehicles. Our research focuses on improving vehicle security by using a type of artificial intelligence called a deep learning model to detect these kinds of attacks. We designed a lightweight model based on a special type of neural network called a 1D Convolutional Neural Network (1D CNN). This model is simple enough to be deployed in real-time on the small computers that cars use. We trained and tested our model using the ROAD dataset, which includes real-world, physically verified attack scenarios that a vehicle might face. This makes our research especially relevant to real-world vehicle security challenges. Our model achieved a detection accuracy of 97.85% on the ROAD dataset and performed exceptionally well on other benchmark datasets, showing that it can generalize to different types of attacks. We also optimized our model using techniques like class weighting and hyperparameter tuning to handle the challenges of imbalanced attack data. This work demonstrates how a carefully designed deep learning model can provide robust, real-time protection for connected vehicles against cyberattacks.

Featured Image

Why is it important?

The security of connected vehicles is a growing concern, as real-world cyberattacks can compromise passenger safety. Most existing security solutions are either too complex for real-time use or tested only on simple, simulated attacks. This work stands out because it uses a lightweight deep learning model trained on the ROAD dataset, which contains physically verified, sophisticated attack scenarios. This makes our model more realistic and reliable for detecting threats in actual vehicles. By demonstrating high accuracy and low false alarms even on challenging datasets, this research provides a crucial step toward real-time, practical intrusion detection in modern cars, enhancing safety and trust in connected vehicle technology.

Perspectives