Evaluation of the Driving Behaviour Models with Newly Collected Data

Driving behaviour analysis is an actual research subject as cities and, more especially, roads become full of cars and ride-sharing applications are more and more popular. Interactive intelligent systems have started to become more and more part of our lives being used in our homes, cars, at school, at shopping, and almost everywhere. The driving behaviour system presented in this chapter is one of those intelligent systems that use AI to detect driving styles seamlessly and to provide feedback for both the driver and the driver’s employer. In this chapter, we present a new dataset along with its design and evaluation based on a set of experiments for estimating driving behaviour. Our method aims to provide a simple-to-use and efficient solution targeting a wide variety of scenarios for collecting data and providing results. We consider that an easy-to-use application which doesn’t get the user involved too much is the key to a usable car application because the user should focus mainly on driving. The method used for building the dataset is based on a mobile application which collects gyroscope and accelerometer data from a smartphone’s integrated sensors. The experiments are conducted using both machine learning and deep learning building models trained on different data configurations. Our experiments present a good and robust accuracy on both the first set of datasets and even after adding a new one. The results show that the application can be used for the estimation of specific driving behaviour and can be integrated into a larger system for further usage.