Comparative Study of Five Classifiers for Predicting Power Conversion Efficiency in Perovskite Solar Cells

Abstract

Perovskite solar cells are promising candidates for low-cost and efficient photovoltaic devices. However, the
experimental design based on trial and error consumes lots of time and money. Machine learning (ML) techniques
can predict perovskite material design and quickly optimize material fabrication parameters and device structure.
In this work, we compare machine learning models to predict the power conversion efficiency of perovskite solar
cells. A 60-point experimental data set with various experimental conditions and device structures are used to
train and test machine learning (ML) models. We trained each of the five classifiers on the 60 data points using 10-
fold cross-validation. We split each dataset into ten equal-sized folds, trained the classifier on 9 -folds, and tested
it on the remaining fold. We repeat this process ten times so that each fold is used for testing once. The machine
learning algorithms or classifiers discussed in this paper include ZeroR, Linear Regression, Gaussian Processes,
Random Tree and Random Forest. By comparing the performance of five different classifiers on 60 data points
of perovskite solar cell data, we can better understand which methods are most effective for predicting PCE.
This article also emphasizes the application of symbolic regression and machine learning to design robust and
effective halide perovskite materials. Additionally, it acts as a foundation for additional experimental perovskite
material optimization. This work has been performed using Weka the machine learning software.