The importance of signal pre-processing for machine learning: The influence of Data scaling in a driver identity classification
Najmeddine Abdennour, Tarek Ouni, Nader Ben Amor
Abstract
Machine Learning (ML) and Deep Learning (DL) algorithms have overtaken the attention of the scientific community for their important capabilities and their over the top results. However, the excessive focus on hyperparameters and the model’s architectures made the pre-processing step often neglected. In spite of its importance, it represented a weak point for most of the machine learning applications as well as a blind spot in many research studies. In this paper, we will demonstrate through a CAN-Bus vehicle data-based driver identification case study, the importance of testing the use of different methods of data scaling and normalization while demonstrating their role in improving the performance of several Machine Learning algorithms.