A data‐centric investigation on the challenges of machine learning methods for bridging life cycle inventory data gaps
Bibliographic record
Abstract
Life cycle assessment (LCA) is a systematic approach to quantify the environmental impacts of a product system from its entire life cycle. Despite its wide use in assessing mature technologies, the inventory data gap has been a fundamental challenge that limits the application of LCA to emerging new processes. Machine learning (ML) methods are among the possible solutions that can mitigate these data gaps in an automated and scalable way. Nonetheless, the performance of existing ML methods is unstable which limits the trustworthiness and generalizability of the models. In this study, we conducted a data-centric investigation to delineate the causes of the unstable performance using a similarity-based ML framework based on Ecoinvent 3.1 unit process (UPR) database. We found that the pattern of imbalance in the data for method development, manifest by the substantial differences in (1) flow and process availability and (2) the order of magnitude of their values, is a major cause of the unstable performance. We also identified the causes due to the challenges with ML method development workflow, particularly, the steps of data preprocessing, and ML model training (e.g., randomness in train-test data splits). In addition, we also tested the proposed ML method on the U.S. Life Cycle Inventory Database, where we observed that the generalizability of the method was highly influenced by the database size of the application. To address these issues, we proposed that further research should focus on reducing the barriers in database integration such that both the size and balance of the data for ML method development can be improved. Supplementary Information: The online version of this article (doi:10.1111/jiec.70022) contains supplementary material, which is available to authorized users.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.059 | 0.146 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.003 | 0.004 |
| Science and technology studies | 0.001 | 0.002 |
| Scholarly communication | 0.005 | 0.008 |
| Open science | 0.003 | 0.002 |
| Research integrity | 0.002 | 0.005 |
| Insufficient payload (model declined to judge) | 0.001 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".