An Improved Prediction of Transparent Conductor Formation Energy using PyCaret: An Open-Source Machine Learning Library

Abstract

Designing innovative materials is necessary to solve vital energy, health, environmental, social, and economic challenges. Transparent conductors are compounds that combine low absorption visible range and good electrical conductivity, which are essential properties for conductors. Technological devices such as photovoltaic cells, transistors, photovoltaic cells and sensors majorly rely on combining the two properties due to their relevancy in an optoelectronic application. Meanwhile, fewer compounds exhibit both outstanding conductivity and transparency suitable for their application in transparent conducting materials. Kaggle hosted an open big-data competition organized by novel material discovery (NOMAD) to address the importance of finding new material with the ideal functionality. The competition was organized to identify the best machine learning (ML) to predict formation enthalpy (indicating stability) for 3000  (AlxGaylnz)2NO3Ncompounds datasets; where x, y, and z can vary from the constraints x+y+z=1. Here we present a prediction using an open-source machine learning library in Python called PyCaret to summarise top-ranked ML algorithms. The gradient boosting regressor (GBR) model performed best with MAE 0.0281, MSE 0.0018 and R2 0.84. The research shows that Machine learning can significantly accelerate the discovery and optimization of materials while reducing cost of computation and required time. Low code tools like PyCaret were used to enhance the machine learning applications in materials science, paving way for more efficient materials discovery processes.