Adaptive Cloud-Based Big Data Analytics Model for Sustainable Supply Chain Management

Abstract

Freight decisions are simultaneously cost decisions, service decisions and disclosable emission decisions, yet in almost every deployed analytics stack the optimiser sees only the first two. We present a cloud-native architecture that moves the emission calculation onto the operational data path, so that a carbon figure can shape a fulfilment decision rather than document one after the fact. The design combines a streaming emission operator conformant with ISO 14083, a leakage-pruned predictive layer, an online assignment rule whose carbon budget is enforced by a Lagrangian multiplier, and an autoscaler whose set point derives from a queueing-stability condition. We evaluate on the DataCo Global supply chain release: 180,501 usable order lines, geocoded to city level for 99.12% of records, with modal emission factors taken from the EPA GHG Emission Factors Hub. Three findings stand out. Cost and carbon turn out to be aligned rather than opposed in this network, because surface transport is both the cheapest and the cleanest option, so the binding trade-off is service level against emissions rather than money against emissions. Against a service-oriented baseline the controller cuts transport emissions by 35.0% for 1.54 days of additional mean lead time, holding SLA attainment at 100%, and the reduction is sign-stable across every declared assumption we sweep. The platform’s own carbon is negligible at this scale, 0.047% of logistics emissions, with the incremental cost of running the controller amounting to 0.0039% of what it saves. Finally, and independently of the framework, models trained on the unpruned DataCo frame achieve perfect separation because four columns are realised only after delivery; removing them lowers PR-AUC by 0.171 (95% CI 0.166 to 0.176).