A Data Mining Based Method for Route and Freight Estimation

We present a method, which makes use of historical vehicle data and current vehicle observations in order to estimate 1) the route a vehicle has used and 2) the freight the vehicle carried along the estimated route. The method includes a learning phase and an estimation phase. In the learning phase, historical data about the movement of a vehicle and of the consignments allocated to the vehicle are used in order to build estimation models: one for route choice and one for freight allocation. In the estimation phase, the generated estimation models are used together with a sequence of observed positions for the vehicle as input in order to generate route and freight estimates. We have partly evaluated our method in an experimental study involving a medium-size Swedish transport operator. The results of the study indicate that supervised learning, in particular the algorithm Naive Bayes Multinomial Updatable, shows good route estimation performance even when significant amount of information about where the vehicle has traveled is missing. For the freight estimation, we used a method based on averaging the consignments on the historical known trips for the estimated route. We argue that the proposed method might contribute to building improved knowledge, e.g., in national road administrations, on the movement of trucks and freight.