Manifest Monitoring Model as Support for Customs Risk Management: Evidence from Taiwan

Abstract

This study constructs a feature prediction model in border smuggling inspection using the Artificial Neural Networks technique. We extracted six input variables relevant to smuggling through an iterated learning process on 176,869 import manifests. By adopting logistic regression, we also revealed that the six identified factors are significantly related to the occurrence of smuggling. We further used the six seizure factors to predict smuggling in 2019 and obtained an average accuracy rate of nearly 83 per cent. The accuracy rate was much higher around Fridays and holidays. The results will help provide cost-effective screening during customs clearance inspection and effectively manage border risks.