Analysis of photoelectrochemical water splitting using machine learning

In this study, an extensive dataset containing 10,560 data points obtained from 584 experiments in 180 articles for photoelectrochemical water splitting over n-type semiconductors was analyzed using machine learning techniques. After the pre- analysis of the dataset using simple descriptive statistics, association rule mining (ARM), random forest (RF) and decision tree (DT) were utilized to identify the patterns in the data and establish relations between photocurrent density and 33 descriptors including electrode materials and synthesis methods as well as the properties of irradiation and electrolyte solution. ARM was successfully employed to identify the critical foctors for band gap and photocurrent density. DT model for the band gap was quite successful (with the overall training and testing accuracies of 78% and 72% respectively) while the model was less accurate for the photocurrent density even though it is still notable (training accuracy = 61%; testing accuracy = 54%). Predictive model developed by random forest for the band gap of the electrode was also remarkably good with the root mean square error of validation and testing 0.24 and 0.27 respectively. (C) 2022 Hydrogen Energy Publications LLC. Published by Elsevier Ltd. All rights reserved.

Automated summary

In this study, machine learning techniques were used to analyze an extensive dataset on photoelectrochemical water splitting over n-type semiconductors. The results showed that association rule mining, decision tree, and random forest could identify patterns and establish relations between photocurrent density and various descriptors.