A Recursive Data-driven Model for Traffic Flow Predictions for Locations with Faulty Sensors

Traffic flow prediction, as an integral part of intelligent transportation systems, is carried out by using data from the traffic sensors in the network. In a typical traffic prediction for a specific location, the most essential data has traditionally been the recent traffic flow records obtained from that same location. However, sensors are prone to failure, and as a result recent traffic measurements for the location of interest may be unavailable. In this work, we propose a model that predicts the traffic flow for locations with faulty sensors, solely by using traffic measurements from neighboring sensors. We use an online recursive regression approach to train the predictive model. We demonstrate the efficiency and accuracy of the proposed methodology using sensor data from California freeways. The results show that the developed model successfully predicts traffic flow with more than 95% accuracy on average.