Determining the Tiers of a Supply Chain Using Machine Learning Algorithms

Kyoung Jong Park

doi:10.3390/sym13101934

Abstract

Companies in the same supply chain influence each other, so sharing information enables more efficient supply chain management. An efficient supply chain must have a symmetry of information between participating entities, but in reality, the information is asymmetric, causing problems. The sustainability of the supply chain continues to be threatened because companies are reluctant to disclose information to others. If companies participating in the supply chain do not disclose accurate information, the next best way to improve the sustainability of the supply chain is to use data from the supply chain to determine each enterprise’s information. This study takes data from the supply chain and then uses machine learning algorithms to find which enterprise the data refer to when new data from unknown sources arise. The machine learning algorithms used are logistic regression, random forest, naive Bayes, decision tree, support vector machine, k-nearest neighbor, and multi-layer perceptron. Indicators for evaluating the performance of multi-class classification machine learning methods are accuracy, confusion matrix, precision, recall, and F1-score. The experimental results showed that LR and MLP accurately predicted companies (tiers), but NB, DT, RF, SVM, and K-NN did not accurately predict companies. In addition, the performance similarity of machine learning algorithms through experiments was classified into LR and MLP groups, NB and DT groups, and RF, SVM, and K-NN groups.

Highlights

The supply chain model targeted targeted in this this study study consists consists of ofaamanufacturer, manufacturer, distributor, distributor, wholesaler, and and retailer, retailer,as asshown shownin inFigure
In this study, assuming that customer demand follows the AR(1) model, a machine learning algorithm was used to determine the tier that the order data of unknown origin refer to in the supply chain model consisting of manufacturer, distributor, wholesaler, and retailer
logistic regression (LR) and multi-layer perceptron (MLP) have the highest accuracy of 100%, random forest (RF) is

Summary

Introduction

Because various companies participate in the supply chain and the environments of these companies are different, complex and diverse situations occur in the supply chain. There are many ways to alleviate problems such as the bullwhip effect occurring in the supply chain, the most representative method is to activate information sharing among companies participating in the supply chain [1]. Information sharing in the supply chain is very important, and information sharing between companies and information sharing with customers has a positive relationship with the performance of the supply chain [2]. The basic principle of supply chain management is to improve the efficiency of the supply chain by maintaining sustainable relationships between companies, and the quality of information shared between companies has a positive effect on the sustainability of the supply chain regardless of the nationality of the company [4]. The supply chain model needs to be symmetrically and harmoniously constructed by all participating entities to pursue the overall optimization of the supply chain

Methods

Results

Conclusion