Illegal timber trade has adverse ecological as well as economic effects. Trafficking in wildlife, plant, and plant products (such as paper, timber) is outlawed under the US Lacey Act (originally introduced in 1990 and amended as recently as 2008). In addition to outlawing the trading of certain categories of products, the law also requires importers to declare the origin country, name, and quantity of products imported in other categories. The Lacey Act is considered a highly effective act, best evidenced by the $13M fine imposed in 2016 by the US Department of Justice on the Virginia company Lumber Liquidators Inc. for importing hardwood manufactured in China from timber illegally logged in Eastern Europe. Agencies tasked with enforcement of laws like the Lacey Act require automated tools to better identify "red flags" for potentially illegal timber shipments, utilizing timber trade data and other publicly available data. Such tools do not exist currently and thus there is a critical need for a better decision support system for assessing trade data for potential anomalies.
We present a machine learning solution to flag instances of potentially suspicious records from trade data. Our system merges disparate data sources such as IUCN Redlist, CITES, and trade data from Panjiva Inc. We use a deep learning based embedding approach to conduct unsupervised anomaly detection. Our user interface allows the analyst to visualize trade data using Sankey diagrams and inspect identified anomalies. To support user understanding, it allows the display of the anomalous records in the context of nearby records.