Missing Data Estimation for Traffic Volume by Searching an Optimum Closed Cut in Urban Networks

Traffic data imputation has drawn significant attention from both academia and industry because traffic data often suffer from data missing problems, caused by temporary deployment of sensors, detector malfunction, and lossy communication systems. To fully exploit the spatial-temporal correlation and road topological information in an urban traffic network, we propose an optimum closed cut (OCC)-based spatio-temporal imputation technique, which is implemented in two stages: a) employing graph theory to search the OCC in the road network, for which the traffic on roads intersected by the closed cut has the maximum correlation with that on the target road while minimizing the number of intersected roads; and b) estimating the missing data on the target road using OCC-based Kriging estimator, incorporating both the road topological information and flow conservation law to improve the estimation accuracy. Experimental results using traffic data collected on real roads indicate that the OCC search algorithm can effectively capture the optimum set of neighboring sensors. An OCC-based estimator can provide more accurate imputation results compared with nearest historical average and correlative ${k}$ -NN ( ${k}$ -nearest neighbors) methods. The road topological information and flow conservation law can be explored to further improve the estimation performance while reducing the number of sensors involved in the data imputation, hence improving the computational efficiency.