driver scorecard and clustering models

This project is asscociated with my academic research paper (to be published).

Quantitative and qualitative estimation of driving style help promote safe and eco-driving. A numerical mapping or a categorical labeling of driving skill level, helps identify from most aggressive to extremely mild drivers for fleet management. With a large scale of aggressive driving events collected via connected trucks, an unsupervised framework is built upon statistics described by those events, and has the ability to grade a driving trip by any driver through objective comparison criteria and to train a better clustering model to classify drivers by their style. The effectiveness of the approach is assessed with an experimental campaign carried out on synthetic and real-world data. Results show that the quantitative part is able to produce interpretable and Normal-like scorecard and the qualitative part help build a better driving style classification model with better driver behavior representation.

Data and Source codes will be availabe here in this project after paper published!

Note: all code and data should not being commercially used