Filling Missing Values in Spatial-temporal Data Collected from Traffic Sensors

Intelligent transportation systems (ITS) are critical to any smart city strategy. They are used to optimize the flow of urban traffic which in turn leads to a reduction in time spent traveling. In order for ITS to work properly, sensors that collect real-time traffic flow information from streets and highways are required so the ITS can know the current state of the traffic. However, such sensors are prone to failures and network faults. This poses a serious hindrance when performing data analysis and knowledge extraction on sensor data due to the fact that such data is composed of noisy and missing values. In this work, we benchmark several deep learning based methods for filling missing values in a dataset collected from 2013 to 2015 in the city of Oporto, Portugal. The dataset is composed of readings of 26 sensors that measure traffic information in 5 minute intervals. Around 12% of all values are missing.