Using machine learning for automated assessment of misclassification of goods for fraud detection

Abstract

The paper is devoted to providing automated solutions to an actual problem of misclassification of goods in cross-border trade. In this paper, we introduce a hybrid approach to Harmonized System (HS) code assessment that combines the knowledge derived from textual descriptions of products, assigned to them HS codes and taxonomy of HS codes nomenclature. We use machine learning for providing HS code’s predictions and recommendations on the basis of a model learned from the textual descriptions of the products. In order to perform an assessment of misclassification of goods we present a novel combined similarity measure based on cosine similarity of texts and semantic similarity of HS codes based on HS code taxonomy (ontology). The method is evaluated on the real open source data set of Bill of Lading Summary 2017 [1] using Gensim Python library [4].