Computational estimation of nano-photocatalyst activity: feasibility of kernel based learning machines

The computational prediction of the nano-photocatalyst activity may significantly speed up the optimization and discovery of novel photocatalysts. In this study, we assess the feasibility of kernel based learning machines to estimate the photocatalyst activity using high-throughput screening data. Nanoparticular anatase-based photocatalyst specimens were characterized in a two dimensional feature space and their activity in emulated daylight was determined in a high-throughput screening procedure. Using this data, a kernel based support vector machine (SVM) was applied to model the relation between the feature space and the activity. After the learning, our scheme provided reasonable estimations of the activity of independent test specimens. It is concluded that kernel based SVMs are feasible for the estimation of photocatalyst activity.