What we did
Generating "ground truth" data through mathematical modeling
We drew from several sources, including a peer-reviewed journal1, to create a proprietary generative model which creates drivers and environments at random and models their interactions to calculate a relative risk score. The risk score is more advanced than existing models as it factors not only the telematic data of the driver or the area in which they drive, but uses complex mathematical models to capture the nonlinear interactions between different variables.
Machine learning as a praxis for capturing the complexity of data
We used deep learning through the keras framework to construct a three-layer deep feedforward perceptron. The perceptron was trained over 10000 generated training points of drivers in different situations and successfully calculated the risk score with over 99.5% accuracy. This simple demonstration is a promising first step which suggests that machine learning, powered by telemetric data, can be used as a holistic driving risk assessment which is more powerful than existing methods and more capable of capturing complex relationships within the data. We are excited to see how this can generalize to real world data.
Helping users minimize risks with a user-friendly dashboard
We built an intuitive dashboard that shows the driver everything they need on what drives their risk up. Whether it's road conditions, weather, or sharp turns, the driver will know immediately what they can fix. This is combined with an interactive map where drivers can pinpoint where exactly on their route they made mistakes and what they were.
- So, B.; Boucher, J.-P.; Valdez, E.A. Synthetic Dataset Generation of Driver Telematics. Risks 2021, 9, 58. https://doi.org/10.3390/risks9040058
Log in or sign up for Devpost to join the conversation.