A General Spatiotemporal Imputation Framework for Missing Sensor Data

Many applications from precision agriculture, environmental monitoring and transportation networks rely on data collected across space and time over a large geographic area. Missing data can be a significant issue in these spatiotemporal databases, as it can reduce the accuracy of downstream data analysis, inferencing and control algorithms. Data imputation or the estimation of missing data can help fill these gaps by utilizing inherent spatial relationships and temporal patterns. However, existing approaches for estimating this missing information do not effectively capture all dimensions of the spatiotemporal data structure, resulting in erroneous predictions and poor performance. In this paper, we introduce a general framework that leverages a spatiotemporal graph constructed from the sensor network graph and temporal sensor data to capture the joint space-time dependencies. Specifically, we propose a graph neural network-based model in conjunction with a recurrent neural network to impute missing information and demonstrate the effectiveness of our approach for downstream tasks. Experiments on a traffic sensor network reveal enhanced imputation accuracy and up to 69% reduction in mean absolute error and 61% reduction in root mean square error compared to state-of-the-art imputation frameworks.