OPTIMIZING SUSTAINABLE SUPPLY CHAIN EFFICIENCY THROUGH DATA VISUALIZATION AND AI-ENHANCED PREDICTIVE ANALYTICS

Abstract

Sustainable supply-chain management increasingly requires analytical approaches that can translate complex freight data into timely insights for reducing operational inefficiencies and transportation-related greenhouse-gas emissions. This study develops an automated data-analytics framework that integrates freight-data processing, carbon-emission estimation, data visualization, machine learning, and anomaly detection to support sustainability-oriented supply-chain decision-making. A randomly ordered working subset of 250,000 shipment records from the 2022 U.S. Commodity Flow Survey was examined, with 121,801 shipments remaining for the emissions analysis after transportation-mode and data-validity restrictions. Shipment-level greenhouse-gas emissions were estimated using mode-specific factors from the U.S. Environmental Protection Agency's 2025 Greenhouse Gas Emission Factors Hub. A Random Forest classifier was subsequently developed to identify shipments in the upper quartile of estimated carbon emissions using operational characteristics available for decision support. The model achieved an accuracy of 89.7%, recall of 91.5%, F1-score of 81.6%, and ROC-AUC of 0.964. A robustness analysis excluding shipment weight retained strong discriminatory capability, achieving a ROC-AUC of 0.932. Permutation importance identified shipment weight and transportation mode as the most influential predictors of high-carbon risk. Isolation Forest analysis was further applied within transportation modes to identify unusually carbon-inefficient shipment patterns while controlling for inherent differences in modal emission intensity. The results demonstrate how automated visualization and AI-enhanced predictive analytics can support earlier identification of carbon-intensive freight activities and enable stakeholders to concentrate on targeted, data-informed sustainability decisions.