Multivariate anomaly detection for ensuring data quality of dendrometer sensor networks
Abstract
Ensuring the integrity of data from large sensor networks is a challenging task that is relevant in many domains. Precision agriculture is one instance of this challenge, where dendrometer sensors provide data used for plot-specific irrigation decisions, with critical implications for yields and water savings. To aid the identification of malfunctioning dendrometer sensors, we introduce a pipeline for detecting various types of anomalies and investigating their root causes using visual analytics. Our pipeline is unique not only in that it borrows from web technologies to provide interactivity, but also because it incorporates detection algorithms from several fields, such as robust multivariate statistics, unsupervised machine learning, and social-network analysis.