The Role of Data Mining in the Risk Assessment of Customs: (With special reference to The Ethiopian Customs Authority)
No Thumbnail Available
Date
2004-07
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
Customs Organizations are responsible for two Opposing In g yet equally important
responsibilities . These are t e provision of efficient services to traders for the
smooth flow of shipments and the protection of the country from any kind of risk
threats that is associated with interactional trade.
Reform and monetization of custom services through automation and setting
transparent working procedures has resulted in the provision of efficient services .
However addressing the issue of implementing a proper risk management strategy
remain s a challenge
The Ethiopian Customs Authority at present is handling the control of customs risks
using subjective methods. The subjective method of handling risks solely depends
on experts' judgment of selecting shipments for physical examination. The
Subjective method is essential since the knowledge and experience of customs
experts and their observation of the behaviors of intervening agents like traders and
clearing agents is very important. However depending only on subjective analysis
for strategic risk management ha s its shortcomings.
This study was aimed at supporting the current selective physical examination
system o f incoming shipments in the Ethiopian Customs Authority with objective
methods using data mining. The study was conducted through the annals is of
customs fraud cases seized in the past. For this stud y one of the data mining
techniques known as decision tree was employed. The dataset used in the stud y
consisted of 10 364 record s out of which 17 0 cases were fraud cases.
The distribution of the two classes was highly imbalanced. To deal with the class
imbalance pro blend the over- sampling approach was use d. Five experiments we re
conducted by varying the rate of over-sampling. After over- sampling the five
datasets had a ratio of 90: 10, 80:20, 70:30, 60:40 and 50 :50 all non- fraud to fraud.
Using an independent dataset that also contain 2616 non-fraud and 39 fraud cases
all the models generated by the five data sets were validated. The dataset with a
proportion of 70:30 has shown the best result in terms of correctly classifying the
fraud cases. The model correctly classified 22 fraud cases out of the 39 cases.
Combining the subjective method with the subjective methods can improve the
efficiency of risk assessment and selective physical examination of shipments in the
Ethiopian customs authority. Models developed by automatic analysis can also be
used across all the different customs offices in the country consistently.
Description
Keywords
Information Science