Exploring two-dimensional ferromagnetic materials by machine learning

Abstract

In this project, two-dimensional (2D) materials are classified into the categories 2D magnetic materials and 2D non-magnetic materials. The 2D magnetic materials are further classified into two categories, namely 2D ferromagnetic materials and 2D anti-ferromagnetic materials. It is important to identify 2D ferromagnetic materials because they are used in various important applications such as data storage mechanisms and spintronic devices. Raw data from the 2D material databases such as 2DMatpedia was used in the project. Classical machine learning involving algorithms such as decision trees and support vector machines were used to classify the data obtained from the above databases into the respective categories anti-ferromagnetic, ferromagnetic and non-magnetic materials. Techniques such as deep learning and convolutional neural networks were also employed to classify the data. The best results for classification of the two-dimensional data into magnetic and non-magnetic materials were obtained for the classical machine learning algorithm, gradient boost algorithm with an accuracy of 93.37%. The most important features of the materials for determining the magnetic properties were band gap, number of magnetic atoms, maximum ionic character, presence of chromium and presence of manganese in that order. The magnetic materials classified were further classified into ferromagnetic and anti-ferromagnetic materials. The best accuracy for this process was obtained to be 94.44% using the classical machine learning algorithm, random forest classifier. The average ionic character, maximum average ionic character, yang-omega, and band-center were the key features for determining whether a material is ferromagnetic or anti-ferromagnetic. A web interface was deployed for the model and can be found here: https://twodferromagnetism-model.herokuapp.com. The code used for obtaining the results can be found here: https://github.com/RiyaBOT/ME4101A-ShethRiyaNimish.