Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for scoring candidate routes. One of the methods includes obtaining a predictive model trained on training examples from trip log data, wherein each training example has feature values from a particular trip and a value of a dependent variable that represents an outcome of a portion of the particular trip, wherein the features of each particular trip include values obtained from one or more external data feed sources that specify a value of a sensor measurement at a particular point in time during the trip. Sensor values from one or more external data feed sources of a sensor network are received. Feature values are generated using the sensor values received from the one or more external data feed sources. A predicted score is computed for each route using the feature values for the candidate route.