深層学習とサプライチェーンAIを用いた個人・製品レベル炭素フットプリント追跡のフェデレーテッド・フレームワーク
Federated Framework for Individual and Product-Level Carbon Footprint Tracking Using Deep Learning and Supply Chain AI (原題)
DIVYANSHU PRADHAN Divyanshu pradhan, Sanchit Sanchit
🤖 gxceed AI 要約
日本語
インドの家計消費に着目し、個人のライフスタイル炭素と製品別炭素を統合追跡するフェデレーテッド型UCFMポータルを提案。12種の回帰モデルを比較しMLPが最高精度（R²=0.8963）を達成。製品側はLLMによる原材料解析とBi/Cross-Encoderの意味的マッピング、XGBoost補完でR²=0.9410を実現し、ISO 14067準拠の監査可能な算定基盤を構築する。
English
Proposes a federated Unified Carbon Footprint Management portal calibrated for India, pairing a deep-learning lifestyle baseline with a barcode-driven product tracker. Among 12 regression models, an MLP achieves R²=0.8963 for lifestyle carbon, while an LLM-based ingredient parser plus Bi/Cross-Encoder factor mapping and XGBoost imputation reach R²=0.9410 for product-level footprints. Anchored to ISO 14067, it offers an auditable blueprint for a locally calibrated global carbon standard.
Unofficial AI-generated summary based on the public title and abstract. Not an official translation.
📝 gxceed 編集解説 — Why this matters
日本のGX文脈において
SSBJ・Scope 3開示が進む日本では、製品別・消費財レベルの算定精度向上とLLM活用の自動化は、サプライチェーン排出量把握や統合報告書の信頼性向上に直結する示唆を持つ。地域較正の考え方は日本版算定基盤の設計にも応用可能。
In the global GX context
Amid ISSB/CSRD and Scope 3 disclosure pressure, this work demonstrates how LLM parsing and semantic factor mapping can automate product-level carbon accounting with ISO 14067 auditability—directly relevant to global supply-chain emissions data quality and greenwashing risk.
👥 読者別の含意
🔬研究者:AI×炭素会計の手法比較（MLP・XGBoost・LLMマッピング）と地域較正の実証設計を参考にできる。
🏢実務担当者:製品別・消費財レベルのScope 3算定自動化とLLMによる原材料マッピングの実装可能性を示す。
🏛政策担当者:地域較正型の算定標準とフェデレーテッド設計が、国別炭素標準や開示制度設計の参考になる。
📄 Abstract（原文）
Anthropogenic climate change demands a decisive shift from macro-level policy pledges to granular, item-levelconsumer accountability. While how an individual commutes matters, daily commercial consumption—the sheer volumeof retail goods purchased—frequently dominates a household's true environmental ledger. In rapidly industrializingeconomies such as India, currently the fourth-largest emitter of greenhouse gases (GHG), existing tracking tools remainsplit between static lifestyle surveys and proprietary industrial Life Cycle Assessments (LCA). Furthermore, directlyimporting Western carbon calculators into developing nations introduces severe estimation bias due to divergent dietarystaples, coal-heavy electricity grids, and informal supply chains. This study proposes "Enhanced India," a FederatedUnified Carbon Footprint Management (UCFM) portal that pairs a deep learning lifestyle baseline with a real-time, endto-end barcode scanning engine parameterized specifically for the Indian context. Building upon the 10,022-individualKolkata urban-suburban survey, we evaluate 12 statistical and deep learning regression models (LiR, LaR, RR, ER, DT,RF, GB, SVR, KNN, MLP, LSTM, CNN) for the Lifestyle Carbon Model (LCM), confirming that the Multi-LayerPerceptron (MLP) achieves superior accuracy (R2 = 0.8963, MSE = 1.6268). To directly overcome the primary limitationof survey calculators—the unavailability of specific food ingredient and commercial product data—our Dynamic ProductTracker (DPT) combines fine-tuned LLM ingredient parsing, Bi-Encoder/Cross-Encoder semantic factor mappingagainst Indian CEA/NITI Aayog and Ecoinvent databases, and an XGBoost imputation engine with GS1 Digital Linklogistics (R2 = 0.9410, RMSE = 0.218 kgCO2e). Anchored to a deterministic ISO 14067 calculation layer, this federatedframework establishes an auditable blueprint for an Indian-calibrated global carbon standard.Keywords Carbon Footprint Calculation · Federated Framework · Indian Context Calibration · Deep Learning · Hybrid LCAML · GS1 Digital Link · Global Carbon Standard
🔗 Provenance — このレコードを発見したソース
- openalex https://doi.org/10.5281/zenodo.23541827first seen 2026-10-24 04:48:47 · last seen 2026-11-06 05:00:01
🔔 こうした論文の新着を逃したくない方は キーワードアラート に登録（無料・3キーワードまで）。
gxceed は公開メタデータに基づく研究支援データセットです。要約・翻訳・解説は AI 支援で生成されています。 最終的な解釈・検証は利用者が原典資料に基づいて行うことを前提とします。