Development of risk assessment (predictive) model for Selective import goods examination in Indian customs and central excise department

Abstract

The pressure to handle large increase in trade with available scarce resources in the last decade of the twentieth century, prompted the Indian Customs authorities to change over from the manual processing to computer based electronic processing of customs documents. This led to the introduction of electronic data interchange system in Indian customs (ICES) in late 1990 s. However, the goods examination process under the ICES still remained manual and thus time consuming. In this backdrop, the current thesis suggests that predictive model can be learnt from the past transaction data stored in the EDI databases. Such models can thus predict the likelihood of duty short declaration in the live declaration in an online mode. Therefore, a lot of precious examination effort can be saved by physically examining only those goods which are covered by the declarations identified as fraudulent by this predictive model. The present work illustrates the selection and preparation of relevant data and the development of predictive model by application of classification tree data mining algorithm. The customs data poses the typical problems of skewed dataset, dissimilar training and application dataset and the variable error of misclassification for each case. These problems have been handled by using various data level and algorithm level interventions like under and over sampling, subdivision of target variables, attachment of differential error weights, tree pruning and use of two different algorithms in succession to finally achieve the maximum possible desirable duty short declaration detections with reasonably low examination effort. The predictive models developed by these hybrid applications of these interventions were able to detect over 90% of the total duty short declaration detections with mere 30% of the original examination effort. In addition, the tree structure and significant decision rules of these predictive models throw useful insight in the patterns in the transaction data that can be utilized for making better informed policy decisions in Indian Customs.