Supply Chain Data Analytics for Predicting Delivery Risks Using Machine Learning

Abstract

The most difficult predictive challenge in supply chain disruption management is order delivery delay. Identifying the risk in delivering an order in the scheduled time will help the company to focus on the prioritized orders to mitigate the disruption before its occurrence. This research presents a machine learning-based predictive model for delivery risk prediction of different product orders. The proposed approach deals with an imbalanced class problem, where the frequency of orders which have the delivery risk is rare when compared to the orders that do not. The Area Under the Curve (AUC) score is the selected performance metric for the proposed risk prediction problem. With a comparative analysis, it is found that the Random Forest model in Synthetic Minority Over-sampling Technique (SMOTE) with the Tomek link gives a better performance with an AUC score of 0.80. It is also found that the Random Forest model performs better in SMOTE and SMOTE Tomek oversampling methods, whereas K-Nearest Neighbour (KNN) performs well in the random oversampling technique.