Feature Ranking Methods for Classifying Vehicle Motion and Immobility States

In this paper the algorithm of a speed sensor is proposed on the signals of the only, arbitrarily oriented relative to the axes of the car 3-axis MEMS accelerometer. The speed sensor autonomously determines the transition time from immobility to moving with the help of machine learning algorithm. When the vehicle is started, the speed is calculated by integrating acceleration along the longitudinal axis of the vehicle using the compensation of static acceleration from the road slope. The complete vector is formed from 14 features obtained by static acceleration processing in a sliding window. It is shown that the degree of informativity of any feature varies depending on the accelerometer operating conditions. The ranking issues to identify the three most informative features used at the stage of training and operation of the algorithm for classification of car motion/non-motion states are considered.