Predicting network flows from speeds using open data and transfer learning

Traffic flow/volume data are commonly used to calibrate and validate traffic simulation models. However, these generally obtained from stationary sensors (e.g. loop detectors), which expensive install maintain cover a small number of locations in the transport network. On other hand, Floating Car Data (FCD) readily available at network level, usually sample vehicles. We present an indirect flow estimation approach using transfer learning address scarcity model generalization across cities. Using two cities (Paris Madrid) as study areas, we demonstrate only exogenous features for prediction, mirroring limited predictive without past link flows. Subsequently, use pre-trained on Paris city test Madrid city, investigate scenarios successful learning. Overall, training set must adequately capture flow-speed relationship estimation. Transfer is beneficial when target task minimal, case transferred models outperform newly trained scratch. real-world publicly data, our can help scale smaller dataset larger