Investigating the impact of pretreatment strategies on photocatalyst for accurate CO2RR productivity quantification: A machine learning approach

[Display omitted] •Light irradiation is the most efficient pretreatment way for TiO2 photocatalyst.•The contamination mechanisms are deduced for four pretreatment techniques.•Machine learning strategy helps us to quantify the HCOOH productivity bias. The photocatalytic carbon dioxide reduction reaction (CO2RR) process is one of the most attractive approaches to mitigate the energy crisis by producing a series of fuels such as methane, methanol, formic acid, acetic acid, etc. Unfortunately, these organic products sometimes happen to be the impurity residuals left on photocatalyst surfaces during large-scale industrial manufacture and artificial contamination during pretreatment processes, which will lead to incorrect quantification of CO2RR productivity or even give an inadequate deduction of the reaction mechanism. Through the machine learning approach (random forest), we further quantified the relative importance of each pretreatment in contributing to the clean TiO2 surfaces, where light irradiation gives the highest contribution (49.7%) among all considered approaches, in contrast to the ultrasound bath cleaning (8.7%). Some suggestions and notes were further provided to point out the limitations and possible improvements for specific techniques. Unexpectedly, even a moderate HCOOH production rate of 100 μmol h−1 g−1 over TiO2-based CO2RR catalysts is speculated to suffer from the non-negligible contamination interference, where a nearly 100% of HCOOH quantification bias is introduced to give the risk to precisely capture the CO2RR productivity feature.