Invisible footprints, visible insights: machine learning reveals Scope 3 emissions - Research article

Aug 5, 2026

Scope 3 greenhouse gas emissions are central to corporate carbon footprints but remain hard to quantify because direct value-chain data are limited. This study builds and compares four machine learning models—K-nearest neighbors, random forest, AdaBoost, and XGBoost—to estimate Scope 3 emissions using readily available financial and sustainability performance indicators. Using 10,449 listed-firm observations (2014–2023) across major industries, the evaluation reports R2, MAPE, and RMSLE. XGBoost delivers the highest accuracy, while random-forest feature importance improves interpretability with only a modest accuracy trade-off.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages13

Reading Time33 min

AccessFree to download

Introduction

Scope 3 emissions challenges

Role of machine learning

Research gaps

Methods

Models compared

Data and evaluation

Results

Model performance comparison

Discussion

Interpretability and key drivers

TYPE Original Research PUBLISHED 09 September 2025 DOI 10.3389/frsus.2025.1649150 OPEN ACCESS EDITED BY Dinesh Kumar, Saveetha University, India REVIEWED BY Pablo Tenoch Rodríguez-González, National Council of Science and Technology (CONACYT), Mexico Lorenzo Zanolo, University of Milano-Bicocca, Italy *CORRESPONDENCE Szu-Yung Wang [edwang92@gmail.com](edwang92@gmail.com) †These authors have contributed equally to this work and share first authorship RECEIVED 18 June 2025 ACCEPTED 25 August 2025 PUBLISHED 09 September 2025 CITATION Wang S-Y and Ye N-Z (2025) Invisible footprints, visible insights: machine learning reveals Scope 3 emissions. Front. Sustain. 6:1649150 . doi: 10.3389/frsus.2025.1649150 COPYRIGHT © 2025 Wang and Ye. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY) . The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms. Invisible footprints, visible insights: machine learning reveals Scope 3 emissions Szu-Yung Wang1*† and Nian-Zu Ye2† 1 Department of International Business, National Taiwan University, Taipei, Taiwan, 2 Department of Accounting, Tamkang University, New Taipei City, Taiwan Introduction: Scope 3 greenhouse gas emissions are critical to firms’ carbon footprints yet are often difficult to quantify due to limited direct data, motivating predictive modeling approaches. Methods: We developed and compared four machine learning algorithms (K-nearest neighbors, random forest, AdaBoost, and XGBoost) to estimate corporate Scope 3 emissions using readily available financial and sustainability performance data. We leverage 10,449 listed firm-level data from 2014 to 2023, covering major industries such as semiconductor, steel, textile, and building materials, evaluating performance of each model by a held-out test set with metrics including R2, mean absolute percentage error (MAPE), and root mean squared logarithmic error (RMSLE) . Results: XGBoost achieved the highest accuracy (R2 = 0. 85, MAPE = 15%, RMSLE = 0. 20), outperforming random forest (R2 = 0. 80, MAPE = 20%) and AdaBoost (R2 = 0.78), while K-NN had the lowest accuracy (R2 = 0.60) . The results demonstrate that ensemble tree-based models substantially improve Scope 3 emission prediction accuracy over simpler models. Discussion: Notably, random forest’s interpretable feature importance provided insight into key emission drivers with only a slight accuracy trade-off, highlighting the balance between predictive accuracy and model interpretability. KEYWORDS Scope 3 emission, carbon accounting, supply chain management, machine learning, AdaBoost, XGBoost, random forest 1 Introduction For most companies, the management of Scope 3 carbon emissions is a daunting and critical challenge. Such emissions cover carbon emissions from all indirect sources throughout the company’s entire value chain, including the activities of upstream suppliers and the downstream product use stage, and often constitute a major part of the company’s total carbon footprint (Khurana et al., 2021; Schmidt et al., 2022) . Compared with Scope 1 and Scope 2 emissions, which are usually directly related to the company’s own operations and are easier to monitor and control, Scope 3 emissions occur outside the company’s operational boundaries, and its management relies on data support from multiple stakeholders such as suppliers, service providers and end users. The diversity of data sources, inconsistent reporting standards and highly decentralized supply chain structures all lead to difficulties and uncertainties in collecting emissions information. In the absence of transparent and consi

FAQs

Why are Scope 3 emissions difficult to quantify?

Scope 3 emissions come from indirect sources across a firm’s value chain, outside operational boundaries. Limited direct data, decentralized reporting, and inconsistent standards across stakeholders make data collection difficult and uncertain.

Which machine learning algorithms are used to estimate Scope 3 emissions?

The study compares K-nearest neighbors, random forest, AdaBoost, and XGBoost. All models estimate Scope 3 emissions from financial and sustainability performance data.

What model performs best in predicting Scope 3 emissions and how is it judged?

XGBoost achieves the highest accuracy with R2 around 0.85, MAPE about 15%, and RMSLE about 0.20. Performance is evaluated using held-out test metrics including R2, MAPE, and RMSLE.

Aug 5, 2026

Upload by Evangeline