Vehicle sensors produce streams of data that are used by the vehicle itself, other vehicles and intelligent transportation systems to provide drivers, fleet managers, and insurance companies with important information to improve safety, decrease travel time and cost.
In this work we utilize the sensor data, combining it with data from other sources, including data from video recorders and HERE aggregated information about road segments. We describe problems and our solutions related to data transformation and predictive modeling related to analysis of drivers' behavior, as well as a process of building and validation of these predictive models.
We enriched vehicle sensor data with HERE road information along the route that the driver took, so that we could get a comprehensive understanding regarding the vehicle behavior in its corresponding context.
We used the data to model driving behavior. Wavelet Analysis were used to estimate optimal parameters. We were able to build models to predict if a person is an aggressive driver based on the data.
We described our approach of data munging, data visualization, feature engineering, feature selection, and modeling.
|