Machine Learning to Detect Abnormal Delivery Performance in Supply Chain Operations

This project focuses on detecting anomalies in the shipment performance of a Business-to-Business supply chain using machine learning. Two models were used for the analysis: Isolation Forest and One-Class SVM. The model training was conducted using the 2024 data to minimize the impact of COVID-19. The data was cleaned and standardized. The key delivery-performance variables were also created to support more accurate anomaly detection. The Isolation Forest achieved an accuracy of approximately 87% with a 5% contamination factor, while the One-Class SVM achieved an accuracy of approximately 82%. Both models identified the Shipping Point as the primary contributor to delays. When the trained models were tested on the 2025 dataset, Isolation Forest returned more consistent results and captured a wider range of anomalies, including Delivery Delay and quantity shortages (partial deliveries), while the One-Class SVM focused more on timing issues. Overall, the study demonstrated that machine learning–based anomaly detection can help organizations identify delivery issues early and enhance shipment performance in B2B operations.