A robust methodology for PEC performance analysis of photoanodes using machine learning and analytical data.

Area of Science:

• Materials Science
• Chemistry
• Energy Harvesting

Background:

• Machine learning (ML) application in experimental science is limited by small datasets.
• Photoanodes are critical for solar water splitting, but their performance is variable.
• Understanding performance variability in materials like hematite and bismuth vanadate is key.

Purpose of the Study:

• To develop a data-driven ML approach for predicting photoanode performance with limited experimental data.
• To identify key descriptors influencing photocurrent in inorganic photodevices.
• To establish a robust methodology for analyzing complex material systems.

Main Methods:

• Applied multiple ML algorithms to predict photocurrent values.
• Incorporated clustering to address multicollinearity in analytical data.
• Utilized Shapley analysis for interpretable identification of performance-influencing factors.

Main Results:

• Achieved prediction accuracy (R^2) over 0.85 on hematite, bismuth vanadate, and tungsten oxide/bismuth vanadate heterojunctions.
• Successfully identified key performance-determining factors for photoanodes.
• Demonstrated superior predictability and factor identification compared to traditional methods.

Conclusions:

• The novel ML methodology effectively predicts photoanode performance using limited experimental data.
• The approach provides clear interpretation of complex material interactions.
• This robust scheme advances research and development of efficient photodevices for energy harvesting.