Comparing different Machine-Learning techniques to predict Vehicles' Positions using the received Signal Strength of periodic messages

In this paper, vehicles use the beacons sent by Road Side Units (RSUs) to predict their positions on a road. The reception power is strongly influenced by the distance between a vehicle and the neighboring RSUs and thus Machine-Learning can be used to predict the position of vehicles between RSUs. We have to assume that the vehicles know their own positions, at least for a given duration, to build the model of the machine-learning algorithm. This position information can be obtained for instance from a GPS. When this information is no longer available, the machine-learning algorithm can be used to predict the vehicles' positions. The vehicles can send a position request to the RSUs which will know the reception power of their beacons and the machine-learning algorithm can respond with the estimated position. In this study, we compare four well-known machine-learning techniques : K Nearest Neighbors (KNN), Neural Network (NN), Random Forest (RF) and Support Vector Machine (SVM). We study these techniques with different assumptions and discuss their respective advantages and drawbacks. Our results show that these four techniques provide very good results in terms of position predictions when the error on the transmission power is small.