Lacking unit process data is a major challenge for developing life cycle inventory (LCI) in life cycle assessment (LCA). Previously, we developed a similarity-based approach to estimate missing unit process data, which works only when less than 5% of the data are missing in a unit process. In this study, we developed a more flexible machine learning model to estimate missing unit process data as a complement to our previous method. In particular, we adopted a decision tree-based supervised learning approach to use an existing unit process dataset (ecoinvent 3.1) to characterize the relationship between the known information (predictors) and the missing one (response). The results show that our model can successfully classify the zero and nonzero flows with a very low misclassification rate (0.79% when 10% of the data are missing). For nonzero flows, the model can accurately estimate their values with an R2 over 0.7 when less than 20% of data are missing in one unit process. Our method can provide important data to complement primary LCI data for LCA studies and demonstrates the promising applications of machine learning techniques in LCA.
Keywords: XGBoost; decision tree; life cycle assessment; life cycle inventory; machine learning; unit process.