Efficient and Effective Anomaly Detection in Autonomous Vehicles: A Combination of Gradient Boosting and ANFIS Algorithms

Abstract

The rise of autonomous vehicles has become a key indicator of smart city development. Unlike traditional cars, which are fully operated by humans, autonomous vehicles rely on sensors to collect data about their surroundings for safe navigation. Due to their reliance on electricity rather than fossil fuels, autonomous cars have a reduced environmental impact in terms of greenhouse gas emissions. However, the susceptibility of autonomous cars to cyberattacks poses a risk to both the vehicles and human lives. Consequently, this study aims to identify and differentiate anomalies in real-time sensor readings of autonomous vehicles. Initially, a fuzzy logic controller with two inputs and one output was fine-tuned to serve as the base controller. Subsequently, data were collected to train the ANFIS-based controllers, each of which was evaluated using three simulations: step response, sine wave response, and random response. The PSO-ANFIS was used to generate an anomaly dataset by introducing artificial false data, and the ensemble model demonstrated exceptional performance, achieving a 99.99% accuracy in classification.