Evaluating Autonomous Vehicle Safety: A Comparative Analysis of ADS and ADAS Incident Data Using Machine Learning Models

This research evaluates the safety performance of Advanced Driver Assistance Systems (ADAS) and Automated Driving Systems (ADS) by training supervised machine learning models on incident report data to predict injury severity. The best results were achieved using a fine-tuned Random Forest model, offering insights to improve autonomous systems and enhance road safety.

This paper focuses on Autonomous Vehicles (AVs), which are rising rapidly in road transportation. However, the reliability of AV systems, namely Advanced Driver Assistance Systems (ADAS) and Au- tomated Driving Systems (ADS), remain vague. Therefore, it is essential that we improve their safety measurements. This study aims to evaluate and compare the safety performance of ADAS and ADS by training supervised machine learning models on incident report data, specifically focusing on predicting injury severity, using numerical and categorical data related to crash conditions and environmental fac- tors. We compared different models, for each fine-tuning their respective hyperparameters. The best per- formance was achieve with Random Forest with hyperparameters max depth = 20, n estimators = 200. This work is expected to provide valuable insights for both the improvement of autonomous systems and overall road safety

Explore More!

Research Paper