Carbon-Aware Multi-Source Data Fusion Model for E-Commerce Demand Forecasting to Reduce Last-Mile Delivery Emissions and Overproduction Waste

• Abiodun Okunola

  Ladoke Akintola University Technology

  Author

The rapid expansion of e-commerce has intensified two interconnected sustainability challenges: last-mile delivery emissions driven by demand uncertainty and overproduction waste resulting from inaccurate forecasting. Existing demand forecasting models prioritize accuracy while neglecting carbon outcomes, and conventional last-mile optimization approaches treat demand as exogenous. This study develops and validates a Carbon-Aware Multi-Source Data Fusion (CA-MSF) model that integrates sales time series, weather parameters, social media sentiment, holiday calendars, and dynamic carbon intensity signals to simultaneously predict demand and inform carbon-optimized fulfillment decisions. Using a design-based research approach combining retrospective e-commerce transaction data (n = 847,392 orders, 2022–2025) with simulation experiments, the CA-MSF model achieved a Mean Absolute Percentage Error (MAPE) of 10.6%, representing an 89.4% accuracy rate in demand forecasting. More critically, carbon-aware demand signals enabled a 22.3% reduction in last-mile emissions through consolidated routing and mode-shifting, while overproduction waste decreased by 18.7% compared to static forecasting baselines. The framework demonstrates that embedding carbon awareness directly into the forecasting layer—rather than treating it as a downstream optimization constraint—produces synergistic gains in both operational efficiency and environmental performance. The study contributes a replicable architecture for carbon-aware demand intelligence and establishes empirical benchmarks for sustainable e-commerce operations.

• pdf

Copyright (c) 2026 Abiodun Okunola (Author)

This work is licensed under a Creative Commons Attribution 4.0 International License.