Scholarlink

<p>Superconductivity, a quantum mechanical marvel in condensed matter systems, continues to challenge fundamental understanding of composition-property relationships in superconducting materials. To address this knowledge gap, we present a machine learning framework analyzing 16,413 superconducting compounds from the <i>SuperCon</i> database. Using only chemical formulas as input, our random forest model achieves 93.5% accuracy and a relative root mean square error of 0.13 in predicting critical temperatures (<i>T</i><sub><i>c</i></sub>) through five-fold cross-validation. Additionally, the methodology demonstrates dual analytical capabilities through material-class-specific regression and <i>T</i><sub><i>c</i></sub>-range classification: Regression models tailored for iron-based (<i>N</i> = 1557), cuprate (<i>N</i> = 4403) and other (<i>N</i> = 6479) superconductors exhibit robust generalizability with R² &gt;0.83 on test sets, while classification models categorizing materials into low- (&lt; 10&#xa0;K), medium- (10–77&#xa0;K), and high-Tc (≥ 77&#xa0;K) groups achieve 91% <i>F</i><sub><i>1</i></sub>-score using random forest classifiers. When deployed for high-throughput screening of 4914 ABO<sub>3</sub>-type perovskites, the optimized model identifies three thermodynamically stable candidates (E<sub>hull</sub> = 0 meV/atom) with predicted <i>T</i><sub><i>c</i></sub> &gt;70&#xa0;K. This data-driven framework not only deciphers composition-property relationships but also establishes an accelerated pathway for targeted discovery of high-temperature superconductors, bridging materials informatics with experimental synthesis priorities.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Discovery of novel High-Tc superconductors via machine learning-based random forest model

  • 期刊论文

  • 发表日期：

    2025年7月10日

查看全文
  • Qi-Xian Wang,
  • Xiang-Fei Yang,
  • Ren-Gang Shi,
  • Yi-Ming Ren,
  • Xiao-Bo Fan,
  • Zhen Qin,
  • Jun Chen

摘要

Superconductivity, a quantum mechanical marvel in condensed matter systems, continues to challenge fundamental understanding of composition-property relationships in superconducting materials. To address this knowledge gap, we present a machine learning framework analyzing 16,413 superconducting compounds from the SuperCon database. Using only chemical formulas as input, our random forest model achieves 93.5% accuracy and a relative root mean square error of 0.13 in predicting critical temperatures (Tc) through five-fold cross-validation. Additionally, the methodology demonstrates dual analytical capabilities through material-class-specific regression and Tc-range classification: Regression models tailored for iron-based (N = 1557), cuprate (N = 4403) and other (N = 6479) superconductors exhibit robust generalizability with R² >0.83 on test sets, while classification models categorizing materials into low- (< 10 K), medium- (10–77 K), and high-Tc (≥ 77 K) groups achieve 91% F1-score using random forest classifiers. When deployed for high-throughput screening of 4914 ABO3-type perovskites, the optimized model identifies three thermodynamically stable candidates (Ehull = 0 meV/atom) with predicted Tc >70 K. This data-driven framework not only deciphers composition-property relationships but also establishes an accelerated pathway for targeted discovery of high-temperature superconductors, bridging materials informatics with experimental synthesis priorities.