Deep Mining of Custom Declaration for Commercial Goods
In our increasingly globalised world, the study of impediments to international trade is of interest to the field of international economics. This paper focuses on the particular problem of speedy and accurate processing of customs declarations. We present a novel use of graph based spreading activation algorithm for the automated processing of customer declarations for commercial goods, based on supervised learning. This method allows us to build recommender systems for use by customs officers, traders, carriers and insurers. We examine the particular use case of the recommendation to assign or not assign an armed escort to a shipping vehicle in cases of elevated risk of theft. In contrast to the usual risk based approach, this algorithm is trained solely on shipment data rather than on traditional risk indicators. This is useful as the recommendation to customs officials can be explained in terms of the make-up of a shipment and can be verified in real-time. The feasibility of the approach was tested by application to 2500 custom records collected during a continuous period of one month at eight border checkpoints between Russian Federation and two EU countries. The algorithm achieved 100% accuracy under experimental conditions.
Custom declarations; data mining; graph-based methods; spreading activation.