Machine learning algorithms for supporting life cycle assessment studies: An analytical review

Nowadays, industries face increasing pressure to enhance their environmental sustainability scores, particularly in reducing carbon footprints. Life Cycle Assessment (LCA) tools are commonly used to evaluate environmental impacts across organizational levels, enabling predictions for potential improvements. But complexity and diversity of factors influencing these assessments make prediction models difficult to build and validate. Machine learning (ML) techniques present viable solutions to these challenges. This study presents a systematic literature review (SLR) of seventy-eight peer reviewed articles, evaluating the performance of different ML models in Life Cycle Assessments applications. An analytical ranking of these models is provided based on their effectiveness for LCA predictions using the Analytical Hierarchy Process (AHP) and Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS). Results indicate that Support Vector Machine (SVM) achieve a score of 0.64...

Nowadays, industries face increasing pressure to enhance their environmental sustainability scores, particularly in reducing carbon footprints. Life Cycle Assessment (LCA) tools are commonly used to evaluate environmental impacts across organizational levels, enabling predictions for potential improvements. But complexity and diversity of factors influencing these assessments make prediction models difficult to build and validate. Machine learning (ML) techniques present viable solutions to these challenges. This study presents a systematic literature review (SLR) of seventy-eight peer reviewed articles, evaluating the performance of different ML models in Life Cycle Assessments applications. An analytical ranking of these models is provided based on their effectiveness for LCA predictions using the Analytical Hierarchy Process (AHP) and Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS). Results indicate that Support Vector Machine (SVM) achieve a score of 0.6412, followed by Extreme Gradient Boosting (XGB) at 0.5811 and Artificial Neural Networks (ANN) at 0.5650, and, positioning them as the most suitable models for LCA studies for prediction application. Random Forest (RF), Decision Trees (DT), and Linear Regression (LR) follow with scores of 0.5353, 0.4776, and 0.4633, respectively, while Adaptive Neuro-Fuzzy Inference System (ANFIS) and Gaussian Process Regression (GPR) rank lowest with scores of 0.4336 and 0.2791. Detailed interpretations and implications of these findings are discussed.