dc.description.abstract |
Trade misinvoicing has become a great problem for all economies especially for the
developing countries like Ethiopia. The importers and exporters are involved in the
misappropriation in invoice settings and prepares trade bill more or less than the
original price of trade invoices. Misappropriation in invoice settings is mainly a problem
for all the countries that are in the process of development. And it is difficult to ascertain
true picture of economy in these circumstances of fake trade records. This study was
conducted to construct model detecting trade misinvoicing at Ethiopian Customs
Commission using machine learning algorithm. Thus, the study has developed model that
detect companies that misinvoicing trade for various reasons. As a result, the study has
used annual report of the organization for two years (2019-2020) for the data set of
31,142 for the transaction made at three routes; Addis Ababa Bole Airport, Kality and
Modjo customs stations. The study data was collected from ASYCUDA database. The
data was used for training and experiment and analyzed by using machine learning
algorithms such as random forest, decision tree, MLP, KNN, and SVC. The predicting
accuracy is highest for random forest classifier from ensemble learner and followed by
decision tree classifier algorithm from tree learner with values of 0.9344 and 0.9094
respectively. On the other hand, lowest accuracy was computed for MLP classifier
algorithm from neural network learner; GaussianNB algorithm from Naive Bayes learner
and SVC algorithm from SVM learner with value of 0.4622, 0.5404 and 0.5502
respectively. The accuracy score of KNeighbors classifier algorithm from neighbor‟s
learner is 0.7464 but this accuracy level is not enough to develop model that significant
number of instances are incorrectly predicted. Based on its performance, the trade
misinvoicing detecting model was developed by using random forest classifier.
Key Words: Under-invoicing, Ethiopia Customs Commission, ASYCUDA, Machine
Learning |
en_US |