MCDM and Machine Learning Approach to Suppliers Management and Selection

Miao Mingyue

Abstract

The current trade war between the United States and China has resulted in high uncertainty in the global supply chains. As a result, many multinational companies now prefer to award only open contracts or separate the whole project to several phases to their key service contractors. This means that contracts and project specifications could be changed or amended based on the actual needs subsequently. The contractors which face more risk now have to manage the uncertainty on the actual needs and may have to procure more products and parts in advance due to the uncertain demand and lead time. The goal of this project is to develop decision support analytics models to help contractors in such situations better manage and select the suppliers and optimize their resources. Based on this goal, the paper studied the principle and method used in supplier selection to form the basic algorithm of analytics model, then extracted the suppliers’ information from semiconductor industries, analyzed the distributed nature of these raw data. The way of extracting data and analyze it is called clustering which is a branch theory in machine learning. To make the model universal, the paper also chose Python language as the programming language to compile this application as Python language is the most popular computing language in the market. Since the model focused on the supplier clustering, validity and feasibility is the big concern to make the model practicable. AHP application was thus involved to do clustering in a different direction apart from the model created in the paper to make sure the model result is valid. The paper used rational design of experiment steps elaborated the project process from setting up supplier selection as a critical issue happened in purchasing department in semiconductor industries to building a suitable clustering model to help anyone involved in purchasing department to do supplier selection easily followed by using existing decision making application to verify the model result to insure the model validity and feasibility. This report also come up with the limitation of current model and ideas of further research direction which can do a good inspiration to practical application in machine learning.