# Data-Mining-Aided-Material Design of Doped LaMnO3 Perovskites with Higher Curie Temperature

## TL;DR

This paper uses machine learning to predict and identify doped LaMnO3 perovskites with higher Curie temperatures for magnetic and spintronic applications.

## Contribution

A predictive framework using support vector regression and high-throughput screening to identify perovskites with elevated Curie temperatures.

## Key findings

- Machine learning model achieved a high correlation (R=0.9111) between predicted and experimental Curie temperatures.
- High-throughput screening identified candidate perovskites with higher Curie temperatures using the OCPMDM platform.
- Web-based infrastructure enables open-access deployment of the validated machine learning model.

## Abstract

The Curie temperature (Tc) of LaMnO3-based perovskites is one of the most important properties associated with their magnetic and spintronic applications. The search for new perovskites with even higher Tc is a challenging problem in material design. Through the systematic optimization of support vector regression (SVR) architecture, we establish a predictive framework for determining the Curie temperature (Tc) of doped LaMnO3 perovskites, leveraging fundamental atomic descriptors. The correlation coefficient (R) between the predicted and experimental Curie temperatures demonstrated high values of 0.9111 when evaluated through the leave-one-out cross-validation (LOOCV) approach, while maintaining a robust correlation of 0.8385 on the independent test set. The subsequent high-throughput screening of perovskite compounds exhibiting higher Curie temperatures was implemented via our online computation platform for materials data mining (OCPMDM), enabling the rapid identification of candidate materials through systematic screening protocols. The findings demonstrate that machine learning exhibits significant efficacy and cost-effectiveness in identifying lanthanum manganite perovskites with elevated Tc, as validated through comparative computational and empirical analyses. Furthermore, a web-based computational infrastructure is implemented for the global dissemination of the predictive framework, enabling the open-access deployment of the validated machine learning model.