Implementation study of a classification model for kinematics and compliance tests in the automotive industry

Identifier:  TFG:9480

Handle: https://hdl.handle.net/20.500.11797/TFG9480

Authors:  Peñalver Hill, Enric

Abstract:
This work studies the feasibility of developing a classification system for experimental vehicle dynamics tests. Synthetic variables are constructed to represent the tests, their relevance is evaluated, and a supervised model (Random Forest) is compared with an unsupervised one (Isolation Forest) to determine the most appropriate training approach. Given the better performance of the supervised models, several algorithms are trained to analyze whether the limitations come from the model or the data. The study detects overfitting, noise, and missing data. As a solution, a graphical tool is developed to improve the labeling process.