A computer-implemented method to determine a device of a driver including receiving, from a first user device of a first user in a vehicle, first telematics data, receiving from a second user device of a second user in the vehicle, second telematics data, determining, using a trained machine learning model, patterns in the first and second telematics data, determining relative positions of the first user device and the second user device by analyzing the telematics data to identify that one of the devices experiences a change in vertical acceleration before the other, determining a delay between the changes in vertical acceleration, determining a front-to-rear positioning of the first user device relative to the second user device based on the delay, determining a predicted position of the first user device relative to the second user device, and identifying the driver of the vehicle based upon the predicted position. Other embodiments are described.