| Abstract |
Mobile Devices, a company specialized in telematics, designs and commercializes devices for vehicles. These devices, connected to the cloud of the company, periodically upload data from hundreds of vehicles sensors with a very low latency. Using these data, Mobile Devices would like to answer at many important questions: Can we model automatically a user driving profile? Can we categorize the different manner of driving type into different classes? How can we detect as soon as possible a sudden change of driving behavior, car breakdown, or an accident? Answering these questions will open a large area of applications to the Mobile Devices company. The objectif of this thesis consists in discovering patterns that discriminate a class (driving behavior, breakdown, …) under three important aspects: (i) the proposition of new pattern languages for having more expressivity, (ii) the proposition of pattern enumeration algorithms (exhaustive search and sampling) and (iii) the use and the adaptation of these methods in a real-time and big-data environment. |
