An Integrated Approach for Supply chain Optimization Using Machine Learning Techniques

Abstract

In the modern world, supply chains completely rely on data to function properly under risk and uncertainty. Supply chain risk optimization is a process that involves identifying, assessing, and managing potential risks within a supply chain network to minimize disruptions. A machine learning analytics model of supply chain risk optimization uses data analytics and machine learning algorithms to understand and assess supply chain risks. Out of many types of risks involved in the supply chain, late delivery risk is the most common, and a lot of attention has been paid by researchers in this regard. The work presented in this thesis utilizes the DataCo Supply Chain dataset. Out of many risks, late delivery and fraud detection are considered in this research work to optimize the risks associated with the supply chain. In total, 15 different machine learning classification algorithms along with two hybrid algorithms are implemented and compared. The better performing hybridized classification algorithm is created in this paper combining the Multi-Layer Perceptron Classifier,Random Forest, and Extra Trees Classifier is put to the test. The hybrid algorithm outperforms all the algorithms and shows an accuracy of 99.45% and 99.15% for late delivery status prediction and fraud detection respectively. In the later part of the thesis, Deep Reinforcement Learning algorithms have been implemented for supply chain pricing policy optimization. The unique factor is that real-time data from an online marketplace in Bangladesh is used in this regard. Deep Q Network and State- Action-Reward-State-Action algorithm have been used, performance-wise Deep Q Network algorithm performed better and it achieved 19% more profit than constant price optimization. The overall work done in this thesis provides a solid foundation of integrated supply chain optimization by which supply chain managers can act proactively and can get benefit.