Sensitivity Analysis of Long Short-Term Memory-Based Neural Network Model for Vehicle Yaw Rate Prediction

In recent years, the application of artificial neural network models has become increasingly widespread in the automotive industry; however, the sensitivity analysis of these models is often neglected. This shortfall poses significant risks in safety-critical applications, where the reliability of models under varying conditions is of critical importance. This study focuses on the sensitivity analysis of a long short-term memory neural network model, previously published by us, designed to predict the future yaw rates of vehicles. Our research aimed to determine the minimum amount of data required for effective model training and to conduct a comprehensive sensitivity analysis, examining the performance and applicability of the trained model under varying tire pressures, different passenger loads, and different passenger configurations. Additionally, we investigated whether the trained model could be applied to other vehicle types. Our results indicated that the vehicle weight distribution was the most influential factor affecting the accuracy of the model. Nonetheless, the model's predictive error remained consistently within the safety thresholds defined by the standards under all tested conditions. Our experiments and analyses were performed using over 7.5 h of data collected under real-world conditions, which will be freely available to the research community.