Application of Matrix Completion Techniques on LCA Data for Different Recycling Scenarios for Parts of Professional Data Centers

When modelling a product by using the LCA methodology, some gaps of knowledge need to be filled in. One solution is to use estimates from available data that could be outdated or only fit roughly the same category. Another solution is to use AI, which generalizes knowledge and can thus provide better estimates. LCA data has a unique structure that certain machine learning algorithms can use to their advantage like the linear dependency between a subset of metrics and scenarios. The research question is how good this generalization works on LCA datasets of limited size with their unique properties. It includes an analysis, which pre-processing techniques that are taking into account the unique structure of LCA data, can improve the prediction performance. Furthermore, the threshold the percentage of missing entries can reach while still ensuring a reasonable performance, is analysed. This paper is a case study that investigates the potential of matrix completion algorithms on LCA data on a small scale using recycling scenarios for parts of professional data centers to derive knowledge for bigger scales.