Intelligent Driver Identification System

Identifying the driver using artificial intelligence can help in many situations, like saving the profile and loading the configurations preferred or helping insurance companies with their plan for each individual driver. Also determining if the person behind the wheel is not the owner/regular driver can help law enforcement agencies to find out faster the theft of a car and it may help them to catch the culprit more efficient. The present work investigates the applicability of neural networks for driver identification in various scenarios: learning all but one driver and see if it classify the left one correctly; learning only one driver and the rest as thieves. The paper analyses the accuracy of neural networks in determining the driver in each scenario, and also checks how the network would react in a stolen vehicle scenario. The data set used for this experiment was collected in 2016, having 94380 records, split between 10 drivers. We obtained promising results in the second and second scenario, the best threshold recorded (99.71%) belonging to the fifth driver, using the configuration with 14 most impactful features and two hidden layers.