Activity coefficients describe the non-ideality of liquid mixtures and are essential for calculating equilibria. The activity coefficients at infinite dilution in binary mixtures are particularly important, as the activity coefficients at finite concentrations in both binary and multicomponent mixtures can be predicted based on their knowledge. Experimental data on this property can be organized in a sparsely populated matrix with rows representing the solutes and columns the solvents. Filling its gaps using predictive methods is essential.
In this thesis, it is shown that matrix completion methods (MCMs) can be applied for this purpose. The novelty of the presented approach compared to traditional prediction methods is that no structural knowledge of the compounds in the considered mixtures is needed. To make temperature-dependent predictions, an MCM is combined with physical knowledge on the temperature dependency of the activity coefficient at infinite dilution. The predictions from this new approach outperform those from the currently best available physical prediction method for activity coefficients at infinite dilution, the modified UNIFAC (Dortmund) method.