Modern highway traffic management systems are greening the ways of mobility by increasing traffic safety, reducing road congestion, pollutant emission, and energy consumption. The efficient operation of the traffic management relies heavily on the quality of the highway sensor data input, which however, is not always reliable. Without appropriate treatment, the potential imperfection in the sensor data may significantly impede the correct traffic estimation and management decision. Most of the previous studies improved the reliability of sensor data by eliminating the sensors that are malfunctioning, at the cost of losing partial information. Instead of fully abandoning the erroneous sensor data, this research attempts to find an approach to process the data, so that the true traffic information can be recovered. The leading-edge statistical learning technologies provide a novel insight into the erroneous data treatment. By incorporating the spatial correlation of traffic counts from sensors in a roadway network, this research will develop a statistical learning based sensor error estimation framework, with the ability in quantifying and correcting the systematic errors of sensors. The corrected sensor data, with more reliable traffic information, contribute to more effective highway traffic management.