Interpretable Machine Learning for Investigating Photoelectrochemical Properties of Cosensitizer-Based CH<sub>3</sub>NH<sub>3</sub>PbI<sub>3</sub>/TiO<sub>2</sub> Films in Water
Lei Zhang, Wenguang Hu, Mu He, Kun Xu, Zhengwei Pan
Abstract
Halide perovskite materials exhibit poor optoelectronic properties under hostile conditions, such as water, therefore interpretable machine learning models of halide perovskite systems under hostile conditions are necessary. In this article, we employ machine learning methods to explore cosensitizer-based halide perovskite films in aqueous solution, with the introduction of multiple types of light-absorbing molecules to modify the perovskite/TiO2 interface. Chemical insights are provided based on the analysis of the molecular descriptors, suggesting the importance of the electrostatic and electrotopological features as well as the chemical compositions and functional groups of the interlayer molecules. Experiments are carried out to validate the machine learning model, and fair agreements between the machine learning model and the experimental results are achieved; this leads to the identification of alternative cosensitizers that offer promising optoelectronic properties in aqueous solution. This article highlights the model interpretability to predict and understand the halide perovskite-based films, and the approach outlined here can be generalized to design other surface systems.