Master's Thesis: Regression Models for Setting Vehicle Attributes
Dec 2019 - Dec 2020
In the early stages of the automotive development process a lot of effort goes into the specification of a consistent set of targets. In this context, a target can be anything from the vehicle’s price to a performance indicator such as top speed. Due to opposing economic and technological ambitions, conflicting vehicle targets are inevitable. This work investigates data-driven models on their potential for supporting the target definition process. The focus is here on vehicle attributes of high relevance to consumers.
Vehicle characteristics that qualify as targets were selected from a database. They act as the supervisor in the respective machine learning problem. The underlying raw data underwent extensive data preparation to meet the requirements of an unbiased, high-quality dataset. Statistical tests were performed to assess the predictive power of single inputs before the multiple regression models were built using a forward selection technique. The output, a mathematical model for predicting a vehicle attribute based on dependent variables, can be further used for calculating all possible solutions. By combining all possible solutions of all models, a solution space could be created that outlines all technological feasible options. This concept was examined with the available models in the sport utility vehicle segment and shows promising results.
