Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Explainable AI for Curie Temperature Prediction in Magnetic Materials

M. Adeel Ajaib, Fariha Nasir, Abdul Rehman

TL;DR

Predicting the Curie temperature $T_c$ of magnetic materials is challenging due to complex chemistry and magnetic interactions. The authors build a prediction pipeline on the NEMAD database, enriching it with composition-based descriptors from $pymatgen$ and Magpie features, and compare multiple models, identifying the Extra Trees regressor as the top performer with $R^2$ up to about $0.87$ on the full data and $0.85$ (±0.01) on a balanced subset. They apply SHAP to reveal physically meaningful drivers such as the mean magnetic moment and Fe fraction, and use $k$-means clustering to uncover chemical subgroups with cluster-specific predictive regimes; excluding poorly modeled clusters improves accuracy to $R^2 \approx 0.85$, MAE about $54$ K, and RMSE about $105$ K. The work demonstrates that explainable AI can enhance both predictive performance and interpretability of compositional factors governing ferromagnetism, with potential extension to other material properties.

Abstract

We explore machine learning techniques for predicting Curie temperatures of magnetic materials using the NEMAD database. By augmenting the dataset with composition-based and domain-aware descriptors, we evaluate the performance of several machine learning models. We find that the Extra Trees Regressor delivers the best performance reaching an R^2 score of up to 0.85 $\pm$ 0.01 (cross-validated) for a balanced dataset. We employ the k-means clustering algorithm to gain insights into the performance of chemically distinct material groups. Furthermore, we perform the SHAP analysis to identify key physicochemical drivers of Curie behavior, such as average atomic number and magnetic moment. By employing explainable AI techniques, this analysis offers insights into the model's predictive behavior, thereby advancing scientific interpretability.

Explainable AI for Curie Temperature Prediction in Magnetic Materials

TL;DR

Predicting the Curie temperature of magnetic materials is challenging due to complex chemistry and magnetic interactions. The authors build a prediction pipeline on the NEMAD database, enriching it with composition-based descriptors from and Magpie features, and compare multiple models, identifying the Extra Trees regressor as the top performer with up to about on the full data and (±0.01) on a balanced subset. They apply SHAP to reveal physically meaningful drivers such as the mean magnetic moment and Fe fraction, and use -means clustering to uncover chemical subgroups with cluster-specific predictive regimes; excluding poorly modeled clusters improves accuracy to , MAE about K, and RMSE about K. The work demonstrates that explainable AI can enhance both predictive performance and interpretability of compositional factors governing ferromagnetism, with potential extension to other material properties.

Abstract

We explore machine learning techniques for predicting Curie temperatures of magnetic materials using the NEMAD database. By augmenting the dataset with composition-based and domain-aware descriptors, we evaluate the performance of several machine learning models. We find that the Extra Trees Regressor delivers the best performance reaching an R^2 score of up to 0.85 0.01 (cross-validated) for a balanced dataset. We employ the k-means clustering algorithm to gain insights into the performance of chemically distinct material groups. Furthermore, we perform the SHAP analysis to identify key physicochemical drivers of Curie behavior, such as average atomic number and magnetic moment. By employing explainable AI techniques, this analysis offers insights into the model's predictive behavior, thereby advancing scientific interpretability.

Paper Structure

This paper contains 7 sections, 16 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Dataset Preparation and Model Evaluation
  3. SHAP Analysis
  4. K-means Clustering Analysis
  5. Cluster-wise SHAP Analysis
  6. Conclusion
  7. Acknowledgments

Figures (16)

  • Figure 2: Feature importance graph for the Extra Trees Algorithm.
  • Figure 4: Elbow plot showing the inertia values for K-means clustering as a function of the number of clusters $k$. A rapid decrease in inertia is observed from $k = 1$ to $k = 3$, followed by a clear flattening beyond $k = 4$, indicating that adding more clusters yields only marginal improvement in explaining the data structure. This behavior supports the choice of $k = 4$ as a balanced and physically meaningful clustering configuration for the Curie temperature dataset.
  • Figure : (a) Main dataset
  • Figure : (a) Cluster-wise elemental composition percentages.
  • Figure : (a) Cluster-wise elemental composition percentages
  • ...and 11 more figures