Machine Learning Approach to Predict the Curie Temperature of Fe- and Pt-Based Alloys

TL;DR

This paper presents a machine learning model, specifically a Voting Ensemble, that accurately predicts the Curie temperature of Fe- and Pt-based alloys using optimized features and Monte Carlo cross-validation, facilitating alloy design.

Contribution

The study introduces an effective machine learning approach with optimal features and ensemble methods for predicting Curie temperatures, enhancing prior experimental and theoretical techniques.

Findings

Voting Ensemble achieves highest prediction accuracy

Monte Carlo cross-validation improves model robustness

Normalized root mean squared error used as performance metric

Abstract

Various techniques can be employed to determine the temperature of magnetic transformation, whether it be the Curie or Neel temperature. The standard procedure typically involves creating alloys with defined compositions and performing measurements experimentally. Alternatively, these temperatures can be predicted based on a material known physical and chemical properties prior to experiments. We identified an optimal feature set and selected the most effective algorithm. Our findings show that the Voting Ensemble model, when combined with Monte Carlo cross-validation, achieves the highest prediction accuracy. The normalized root mean squared error serves as the primary performance metric. For implementation, we utilize the Azure Machine Learning framework for its robust computational and integration capabilities. This approach offers an efficient and reliable strategy for designing and predicting the Curie temperature of ternary alloys. The paper also highlights potential applications of the model and its extensions for other systems.