A vehicle classification algorithm based on telematics data

Abstract

In the thesis, I develop an algorithm to identify the vehicle model from telematics data. By extracting the features from the accelerometer and GPS data, we obtain the classification features, which then goes through a multiclass random forest classifier. We apply this results into problems of driver and vehicle identification. The result shows that, while the algorithm could identify the vehicle models to some extent, the dominating signal comes from driving style, and an approach running purely unsupervised learning is harder to achieve good classification results compared to supervised methods.