Machine learning as a tool for interpreting variables in hydrogen sorption data

To realize the hydrogen economy, it is important to develop methods of storing hydrogen that are cost-effective, safe, and compact. One method of storing hydrogen which shows a lot of potential is through physisorption on high surface area carbon. However, the scientific literature offers conflicting information on which properties of the carbons are important to hydrogen storage and to what extent. This is chiefly because each experimental study on carbon materials only analyzes results on an incredibly small subset of carbon materials. To remedy this, we conducted an integrative data analysis wherein we collected experimental data from many studies in the literature to construct a large dataset. We then used this dataset to develop a machine learning model which can predict the hydrogen adsorption isotherm at 77K based on the porosity and chemical composition of the carbon. By analyzing this model using a post-hoc explanation method called Shapley Additive Explanations, we can analyze the structure-property relationships clearly.