Artificial Intelligence and Life Cycle Assessment: Support for Data Gap Analysis 

Life Cycle Assessment (LCA) is a widely used tool for evaluating environmental impacts, but its application to emerging technologies is often constrained by data gaps particularly in upstream or region-specific processes resulting in increased uncertainty and potential bias. [1, 2] This is especially problematic when LCA results are used for decision-making, product comparisons, or policy development. Properly addressing data gaps using various approaches such as proxy data, expert judgment, or more recently, artificial intelligence assisted estimation methods improve the robustness and credibility of the analysis. [2, 3] Consequently, managing and reducing data gaps is crucial to maintaining the transparency and scientific integrity of LCA studies. [2] 

Approaches to address LCA data gaps 

Various methods are used to address life cycle inventory (LCI) data gaps. Common approaches include using (a) proxy data from similar products or processes, [4] (b) expert judgment and literature review, [5] (c) data extrapolation and Interpolation techniques, [6] (d) hybrid LCA methods, which integrate process-based and economic input-output data, [7] (e) modeling and process simulation through process simulation software (e.g. Aspen Plus®) or theoretical models to generate new data and estimate environmental impacts, [8] (f) empirical modeling, [9] (e) engaging stakeholders to collect primary data to improve data quality and relevance, and more recently (f) artificial intelligence and machine learning which have been explored to predict or estimate unknown data points effectively. [10, 3, 11, 12, 13] 

The literature also highlights other approaches to addressing inventory analysis and data gaps. For instance, the integration of big data with LCA, which enables dynamic and location-specific analysis for a more accurate reflection of real-world conditions, [14] and the adoption of FAIR (Findable, Accessible, Interoperable, and Reusable) data principles, which improve data sharing, transparency, and reusability in LCA studies. [15] 

However, among the abovementioned approaches to addressing data gaps in LCA, this report focuses on artificial intelligence (AI) and machine learning (ML) application due to their growing capability to efficiently handle complex datasets, predict missing values, and enhance data quality. Their ability to automate and improve data gap analysis makes them especially promising tools for advancing the robustness and reliability of LCA inventories. 

Application of AI and ML in LCA; LCI and data gap analysis 

AI and ML powered by advanced algorithms are being able to predict missing data and identify patterns in complex systems, enhancing the accuracy of LCA results. Here some recent publications in terms of application AI and ML in LCA are briefly discussed. 

AI can analyze large datasets to identify the most impactful areas within a product system, helping to refine the scope of an LCA for a more focused and efficient evaluation. [16] Ghoroghi, et al. in their review paper, discussed the challenges of applying ML techniques to support LCA solutions. They noted that ML can be effectively used in optimization tasks within LCA and is particularly powerful when integrated with existing inventory databases, helping to streamline LCA processes across various applications. [17] 

Koyamparambath, et al. showed that AI can be employed to predict the environmental performance of a product or service using data from environmental product declarations (EPD) of construction products. [3] They noted that the advantage of using EPDs lies in their standardized format and predefined rules, i.e., Product Category Rules (PCRs), which makes them easier to implement AI and they are publicly available and can be easily downloaded. On the other hand, the LCA databases of products are not free, they are expensive and not harmonized, along with various methods and assumptions employed in performing an LCA. The data were processed through natural language processing (NLP) “which is then trained to random forest algorithm (RF), an ensemble tree-based machine learning method.” [3] The model can predict impact categories with some accuracy, which the method presents the capability to estimate environmental performance by learning from the results of the previous LCA studies. They considered different product information extracted from EPDs as inputs to the ML algorithm to predict the impact assessment results for a given category. It was noted that the method is meant as a quick-check tool, not a substitute for a full LCA, due to limited EPD availability and current development stage. Meanwhile, the model performance presented dependency on the amount of data available for training. According to the study which employed this methodology, the approach does not replace a detailed LCA, however, can potentially provide quick prediction and assistance to LCA practitioners. [3] In terms of employing artificial neural networks (ANN) or mathematical models, they mentioned that ANNs are computationally intensive, time-consuming to train, and challenging to control in terms of prediction accuracy. In contrast, mathematical and statistics-based algorithms (e.g., multiple linear regression, Bayes classifier, and decision tree regression) have an advantage on controlling the quality of prediction with lower data. However, for the large data, ANN outperformed these mathematical algorithms. Therefore, the choice of a mathematical algorithm over an ANN is based on data availability and the need for greater control over the prediction process. [3] 

Thombre et al., employing data from EPDs of construction products and processing them through NLP, and trained on a RF algorithm, they introduced an ML method based on ensemble trees. While the model can potentially serve as a fast prediction tool, its performance depends on the volume of available training data. Accordingly, the paper mentions that the method does not replace a detailed LCA however, offers quick support for practitioners and verifiers. [18] 

Romeiko et al. studied the predictive accuracies and efficiencies of five distinct supervised ML algorithms, through testing various sample sizes and feature selections. The findings indicate that ML can reliably replicate process-based models to estimate life cycle impacts over large spatial and temporal scales. [19] 

According to a review paper, in terms of the performance of different ML models in LCAs applications, Support Vector Machine (SVM), followed by Extreme Gradient Boosting (XGB), and ANN are the most suitable models for LCA studies for prediction application. While, Random Forest (RF), Decision Trees (DT), Linear Regression (LR), Adaptive Neuro-Fuzzy Inference System (ANFIS), and Gaussian Process Regression (GPR) are ranked in lower performance, respectively. [12] However, they noted that among the various ML techniques, ANNs emerged as the most widely applied, followed by RF and SVM. Accordingly, it is mentioned that ML applications in LCA are mostly in the life cycle inventory (LCI) and life cycle impact assessment (LCIA) stages, where they facilitate predictive modeling based on diverse datasets, including scientific literature, experimental results, and industry databases. While ML enhances accuracy and automation in LCA, challenges like computational demands, data quality, and lack of standardization persist. 

In another study to address the data gap challenges of LCI, both in foreground and background data the large language models (LLMs) were introduced as the promising methodology. However, the authors of the paper acknowledge that data sets containing incorrect information extracted from literature (e.g., incorrect mapping between flow and provider, incorrect value of electricity use for a specific manufacturing step) can compromise the efficacy of applying LLMs to bridge the data gaps for LCI modeling. Also, completely eliminating such incorrect data using automated data curation methods is currently not feasible. For a better results and reduction the incorrect information in the curated data sets in using LLMs applying techniques and best practices developed for the ML field in general, such as thorough data preprocessing and cleaning, using high-quality data sources, implementing robust, human-in-the-loop validation protocols, applying statistical noise reduction methods, and ensuring transparency and explainability are adviced by the paper. [20] 

Neupane et al., in their paper, reviewed 78 peer-reviewed articles on ML application in LCA studies to identify the current trends, commonly used models, models input and outputs, data sources and data sizes, and methodological challenges. [12] Their analysis was chose based on 11 different criteria. The main takeaways from their review include: (a) ML applications in LCA are predominantly utilized in the LCI and life cycle impact assessment (LCIA) stages, (b) among the various ML techniques, ANNs emerged as the most widely applied, followed by RF and SVM. (c) Emerging AI technologies, such as deep learning and LLMs, offer new possibilities to enhance ML applications in LCA. They discuss that while ML enhances accuracy and automation in LCA, issues like high computational cost, poor interpretability, and inconsistent evaluation standards persist highlighting the need for reliable decision-making tools. [12] 

According to a review paper by Romeiko, et al., ML approaches enhance prediction accuracy, facilitate pattern discovery, and improve computational efficiency. [21] However, they mentioned that several areas require further investigation. (1) Ongoing data collection and compilation to support more robust ML and LCA modeling. (2) Future research to provide clear criteria for ML model selection and include thorough uncertainty analyses. (3) Integrating deep learning techniques into LCA to potentially advance LCI and impact assessment. (4) Necessitates of interdisciplinary collaboration to fully integrate ML into LCA in support of sustainable development. [21] 

Conclusion 

Application of AI and ML techniques offer promising applications in addressing data gaps and estimating LCI data, along with predicting characterization factors and impact categories, and enhancing the interpretation step. Although various ML models have been applied for LCA, standardized guidelines and thorough performance evaluation criteria remain absent. Among various ML methods, ANNs have emerged as the most used approach. Using Environmental Product Declarations (EPDs) as input data is a widely adopted approach for implementing machine learning methods. AI and ML methods, while promising, are not yet viable substitutes for comprehensive LCA due to limited data availability (e.g., EPDs) and early-stage development. However, they can potentially offer rapid estimates and support for LCA practitioners. AI/ML can be integrated with LCA at various phases from LCI, LCIA through interpretation to a decision making potentially employing open-source tools. Although this report primarily focuses on the application of AI and ML in LCI and data gap analysis within LCA, Figure (above) presents a conceptual flow diagram highlighting their potential roles across all phases of the LCA process. 

For LCA experts and practitioners, building a foundational understanding of AI and ML is becoming increasingly essential. They need to explore relevant ML applications in LCA such as data gap filing, environmental impact prediction, and uncertainty analysis. One of the most effective ways to get started is by initiating interdisciplinary projects that combine LCA expertise with ML techniques. A key skill is the ability to frame LCA challenges into ML problems, such as classification or prediction tasks. Utilizing open-source tools like openLCA or database alongside Python can provide a practical entry point for applying ML in LCA work. Additionally, collaborating with AI/ML professionals is crucial for effectively linking environmental assessment with algorithm development.

References

[1] G. Finnvedena, "Recent developments in Life Cycle Assessment," Journal of Environmental Management, vol. 91, pp. 1-21, 2009.

[2] R. K. R. S. I. O. Michael Z. Hauschild, Life Cycle Assessment: Theory and Practice, Springer, 2018.

[3] A. Koyamparambath, "Implementing Artificial Intelligence Techniques to Predict Environmental Impacts: Case of Construction Products," Sustainability, vol. 14, p. 3699, 2022.

[4] M. C. Llorence, "Approaches for Addressing Life Cycle Assessment Data Gaps for Bio-based Products," Journal of Industrial Ecology, vol. 15, no. 5, pp. 707-725, 2011.

[5] L. Greif, "A Knowledge Graph Framework to Support Life Cycle Assessment for Sustainable Decision-Making," Applied Science, vol. 15, p. 175, 2025.

[6] M. Erakca, "Systematic review of scale-up methods for prospective life cycle assessment of emerging technologies," Journal of Cleaner Production, vol. 451, p. 142161, 2024.

[7] R. Hagenaars, "Hybrid LCA for sustainable transitions: principles, applications, and prospects," Renewable and Sustainable Energy Reviews, vol. 212, p. 115443, 2025.

[8] J. Ferdous, "Use of process simulation to obtain life cycle inventory data for LCA: A systematic review," Cleaner Environmental Systems, vol. 14, p. 100215, 2024.

[9] Y. Y. Q. T. Shiva Zargar, "A review of inventory modeling methods for missing data in life cycle assessment," Journal of Industrial Ecology, vol. 26, pp. 1676-1689, 2022.

[10] A. Hamdan, "AI and machine learning in climate change research: A review of predictive models and environmental impact," World Journal of Advanced Research and Reviews, vol. 21, pp. 1999-2008, 2024.

[11] M. Zareba, "Machine Learning Techniques for Spatio-Temporal Air Pollution Prediction to Drive Sustainable Urban Development in the Era of Energy and Data Transformation," Energies, vol. 17, p. 2738, 2024.

[12] B. Neupane, "Machine learning algorithms for supporting life cycle assessment studies: An analytical review," Sustainable Production and Consumption, vol. 56, pp. 37-53, 2025.

[13] B. Zhao, "A data-centric investigation on the challenges of machine learning methods for bridging life cycle inventory data gaps," Journal of Industrial Ecology, pp. 1-12, 2025.

[14] J. Li, "Coupling big data and life cycle assessment: A review, recommendations, and prospects," Ecological Indicators, vol. 153, p. 110455, 2023.

[15] A. Ghose, "Can LCA be FAIR? Assessing the status quo and opportunities for FAIR data sharing," The International Journal of Life Cycle Assessment, vol. 29, pp. 733-744, 2024.

[16] H. E. team, "How to apply AI effectively for Life Cycle Assessment," 2024. [Online]. Available: https://hogonext.com/how-to-apply-ai-effectively-for-life-cycle-assessment/?utm_source=chatgpt.com.

[17] A. Ghoroghi, "Advances in application of machine learning to life cycle assessment: a literature review," The International Journal of Life Cycle Assessment , vol. 27, pp. 433-456, 2022.

[18] S. Thombre, "Prediction of Environmental Impacts through Artificial Intelligence Techniques" Material Science and Technology, vol. 23, pp. 406-412, 2024.

[19] X. X. Romeiko, "Comparing Machine Learning Approaches for Predicting Spatially Explicit Life Cycle Global Warming and Eutrophication Impacts from Corn Production," Sustainability, vol. 12, p. 1481, 2020.

[20] Q. Tu, "Mitigating Grand Challenges in Life Cycle Inventory Modeling through the Applications of Large Language Models," Environmental Science and Technology, vol. 58, pp. 19595-19603, 2024.

[21] X. X. Romeiko, "A review of machine learning applications in life cycle assessment studies," Science of The Total Environment, vol. 912, p. 168969, 2024.