Edge AI for Short-Term Delay Prediction in Distributed Logistics Networks (DECLN)

Abstract

Last-mile logistics is becoming so complex that we need to make decisions to adapt to local conditions within a short timeframe, such as traffic congestion, weather, or even driver performance. This paper proposes an Edge AI model for distributed edge-cloud logistics networks (DECLN), deploying very lightweight predictive models on micro-hubs or delivery vehicles rather than relying solely on heavy predictive models in the cloud. Our edge AI models forecast short-term Delay Probability by combining local sensor data (activity reports for traffic congestion, severity reports for weather, estimates of arrival-time variation, and driver scores). Re-calculate Delivery Time in real time using local input data, then re-allocate delivery locations without requiring entirely cloud-based processing. We implement a basic regionalised logistic regression experiment comparing global (cloud-only) models to local (edge-simulated) models. These results, based on our framework, suggest that much of the reported prediction latency will be reduced and bandwidth usage will be lower while improving localised decision-making accuracy. This paper contributes practical methods to improve responsiveness as delivery environments change.