Can Machine Learning Beat Physics at Modeling Car Crashes?
Document Type Dissertation
Dissertation submitted in partial fulfilment of the requirements of Dublin Institute of Technology for the degree of M.Sc. in Computing (Stream), March 2018.
Abstract
This study aimed to look at a traditional method used for measuring the severity and principle direction of force of a car crash and see if it could be improved on using machine learning models. The data used was publicly available from the NHTSA database and included descriptions of the vehicle, test and sensors as well as the ac celerometer data over the period of the crashes. The models built were SVM classifiers and multinomial regression models. Although the SVM and Regression models were build successfully and gave higher levels of accuracy than the momentum models in terms of the severity, the traditional momentum model’s severity results were not sta tistically significant and it was therefore impossible to say the SVM classifier was an improvement using the same data. The principle direction of force was improved on using both a multi-level SVM classifier and a multinomial regression and the results were statistically significant.