Machine Learning Speed Model

Project Brief

City Science required a speed model to populate gaps in the network where speed data isn’t available (e.g. during certain low-volume time periods or along quieter routes). The aim was not to develop a full 4-stage model to examine these effects, but to create a mechanism to forecast the relevant information with minimal time and complexity.


Project Detail

City Science built a network from the ITN, turning it into a graph of nodes and edges. This was linked to the City Science database of Geotagged / GIS-linked predictor variables. A series of additional variables were created based on the properties of network mathematics. The team then selected a range of non-demand predictor variables and applied the Gradient Boosting Tree algorithm in Python.


The model developed uses minimal amounts of additional data, requiring only local network information and speed data to calibrate. This makes it potentially interesting to local authorities since the cost of the model development does not include any new data. The model can be used to quickly evaluate many network options including the removal of roads or the insertion of new links. This can be useful at the option development stage to develop order-of-magnitude estimates of a range of schemes, before using traditional techniques for detailed modelling.
City Science Speed Model of Exeter