With accelerated development of data storage and handling techniques came the exponential growth of the technical advancement of the academic field called 'data mining'. Using several techniques available today, Korea Customs Office have launched a campaign to develop expert knowledge based data mining system that would detect fraudulent activities relating tariff violations.
Using data from Korea Customs Office, a neural network, a decision tree, and a logistic regression based fraud detection system was trained on a large sample of transactions and was tested on a holdout sample that consisted of all of characteristics of population.
A multi-layer perceptron (MLP) network was trained to classify tariffs violations of commerce trading data from Korea Customs Office. A technique based on the probabilistic interpretation of the output of the neural network was used to see if it improved the performance of the MLP given the extent of noise (i.e. inconsistencies) in the original classification model.
We discuss the performance on this data set in terms of detection accuracy among the three techniques being applied. The system has been installed and is currently in use for fraud detection on tariff violation at Korea Customs Office.
