Ibrahim Shujaz. Customs valuation fraud detection based on instance weighted one class support vector machine. Master's Degree(Information Technology). King Mongkut's University of Technology North Bangkok. Central Library. : King Mongkut's University of Technology North Bangkok, 2010.
Customs valuation fraud detection based on instance weighted one class support vector machine
Abstract:
Customs duty is one of the most important sources of revenue for a country,
particularly for less industrialized developing countries. However many unscrupulous
importers, undervalue or overvalue the actual value of goods to evade the duty or for
money laundering, respectively. This fact makes it paramount for customs to find
automatic or semi-automatic solution to detect such fraudulent activities as soon as
they take place, to avoid revenue loss and other damage to the society. Majority of
the customs import transactions are assumed to be in the normal class. But those
transactions might be mislabeled or noisy. In this context, first we proposed an
instance weighting algorithm to compute the relative importance of individual training
data points in the normal class. Then secondly, we reformulate the one class support
vector machine (OCSVM) to incorporate those weights. Hence we called the
reformulated version, Instance Weighted OCSVM (IWOCSVM). We conducted
experiments using a data set from Maldives Custom Service and two benchmark data
sets to compare the OCSVM and the IWOCSVM. Our results show significant
improvement in performance for IWOCSVM.