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.
