A data-driven framework for environmental impact assessment of building materials using environmental product declarations

Material selection in the early design stage shapes a building's environmental performance, yet comparing products across multiple impact categories is difficult. Environmental Product Declarations (EPDs) supply comparable life-cycle data for building products, but existing selection methods depend on expert-assigned weights or fixed thresholds that often determine the outcome. This study presents an exploratory, data-driven framework that classifies external wall materials by relative environmental burden. Ten environmental and technical indicators were compiled from 333 verified EPD records for brick (110), concrete block (119) and wood (104), drawn from 23 European countries, published 2022–2026, declared under EN 15804+A2 at cradle-to-gate (A1-A3). The dataset was divided using a stratified 80:20 train-test split based on material category. Indicators were standardized and aggregated into a composite burden index, which was classified into Low, Medium and High levels using a quantile-based approach, with every data-dependent parameter, including class thresholds, estimated from training data only. Logistic regression, support vector machine and random forest models were tuned identically using repeated stratified 5-fold cross-validation and evaluated on unseen test data. All three models achieved 0.940 accuracy, indicating that the outcome follows the labelling rule rather than the algorithm. A classifier using material identity alone reached a macro-F1 of 0.55 against 0.97 for the indicator-based models, confirming that classification captures indicator patterns, not material category. Durability, embodied energy and fossil global warming potential were the most influential indicators. The resulting classes are relative to the reference dataset, offering practitioners and stakeholders a transparent basis for early-stage screening rather than absolute ratings.