Exploring machine learning models to predict harmonized system code

Abstract

The Harmonized System (HS) Code is widely used across all customs administrations because of the several benefits including a more convenient and easier approach for calculating duties as well as preventing the potential loss of revenue. This paper aims to explore various machine learning models to predict the HS Code based on the customers’ input commodity descriptions. This prediction model helps in reducing the complexity, gaps and many other challenges in using HS Code in any Customs administration. This study follows the Cross-Industry Process for Data Mining methodology which comprises six phases, namely business understanding, data understanding, data preparation, building prediction model, performance evaluation and model deployment. The results of the study indicate that machine learning models are effective tools in predicting HS Code based on user’s inputs. The linear support vector machine model was able to achieve the highest accuracy of 76.3%.