Data mining in photocatalytic water splitting over perovskites literature for higher hydrogen production

A database containing 540 cases from 151 published papers on photocatalytic water splitting over perovskites was constructed and analyzed using data mining tools; the factors leading high hydrogen production were identified by association rule mining while some useful heuristics for the future studies were developed by decision tree analysis. Additionally, the predictive models were developed using random forest regression. In about half of the works, the perovskites were doped by A-site, B-site or both; however, only some portion of doped catalysts had better activity than plain perovskites while doping also improved stability in some cases. The effect of co-catalyst on activity seems to be also irregular; no definitive conclusion could be drawn. The effects of preparation methods on surface area, band gap and crystal structure were noticeable. This is also observed in visible light activity; for example the materials prepared by hydrothermal synthesis method appeared to perform better under visible light. Methanol and other sacrificial agents were used in both UV and visible light tests while inorganic additives have been commonly utilized under visible light. The band gap was found to be highly predictable but it could not be directly linked to the hydrogen production. As the result, although there has been significant progress in the field, the improvement in hydrogen production appeared to be always limited; the sound solutions like ion doping to modify the band gap, use of co-catalyst for charge separation or use of additives as sacrificial agents did not to help as much as desired.