In this paper a driving mode estimation model based in machine learning architecture is presented. With the statistic method, Random Forest, the highest inference of driving variables is determined through the best attributes for a training model based in OBD II data. Engine sensors variables are obtained with the aim of explaining the behavior of the PID signals in relation to the driving mode of a person, according to specific consumption and engine performance, characterizing the signals behavior in relation to the different driving modes. The investigation consists of 4 power tests in the dynamometer bank at 25%, 50%, 75% and 100% throttle valve opening to determine the relationship between engine performance and normal vehicle circulation, through the engine most influential variables like MAP, TPS, VSS, Ax and each the transmission ratio infer in the fuel consumption study and engine performance. In this study Random Forest is used achieving an accuracy rate of 0.98905.