Characterization and modelling of a customs operation
Abstract
Effective risk management is a prerequisite to find an acceptable balance between the objectives of a customs
operation and the streamlined flow of goods. This requires the use of well-designed customs risk management
models that scrutinize all cargo consignments in cyber space based on the analysis of rich data sets that can be
used to accurately determine the risk represented by a cargo consignment without physically stopping it. The use
of such models results in much reduced physical inspections without increasing the risk to customs of either losing
income or allowing the influx of illegal contraband. It therefore represents a much more optimal compromise
between the interests of customs and those of trade, reducing the economic cost to the region and making the
region more attractive to global economic partners. In this paper we utilize different classification techniques to
recognize patterns in electronic data transacted between customs and trade that characterize the risk attributes of
cargo consignments. We subsequently extract models that can be applied in real time to minimize disruption of
trade flows while reducing customs risks to below set thresholds. We quantify the impact of a variety in input
factors and demonstrate how an optimal set of inputs can be selected to arrive at an effective risk management
model. The diagnostic abilities of linear regression, neural network and classification tree techniques to predict
both customs stops and infractions before they occur are compared