Contextual Anomaly Detection in Logistics and Supply Chain Management

The rapid growth of the Internet has transformed e-commerce and logistics, driving a surge in online shopping and a heightened demand for efficient delivery operations. This evolution has introduced new challenges, as operational delays often arise from incorrect planning or unforeseen anomalies, which can originate from various sources, including shipments, carriers, vehicles, employees, or specific tasks. While some anomalies disrupt delivery timelines, others do not, despite exhibiting abnormal job durations. This study introduces a Contextual Anomaly Detection (CAD) framework to identify and analyze all types of anomalies within operational processes, considering the specific contexts of branches, locations (sender and receiver cities), times, job types, and units. By leveraging real-world data, the research highlights the reliability and practical applicability of the findings. A variety of regression models are employed using the contextual and behavioral features from on-time deliveries without outliers to analyze anomalies in job durations during delivery processes, alongside comparisons with unsupervised methods for CAD. The results highlight the critical role of CAD in accurately identifying abnormal patterns and improving the efficiency of operational workflows. This approach provides logistics companies with actionable insights to optimize their processes, address inefficiencies, and enhance overall delivery performance.