Scholarlink

<p>Rare earth-containing magnesium alloys are critical materials in biomedical applications, yet their corrosion performance directly determines service safety. To overcome the time-consuming limitations of traditional experiments and the difficulty in quantifying complex corrosion mechanisms, this study established a machine learning prediction framework using literature-derived alloy compositions and environmental data. Six algorithms, including Random Forest Regressor, Extreme Gradient Boosting, and Support Vector Machine, were rigorously evaluated. Beyond standard grid search, an advanced optimization strategy integrating the Local Outlier Factor method for noise reduction and learning curve analysis was employed to effectively mitigate overfitting. The results indicate that the optimized Random Forest Regressor model achieved the highest accuracy for corrosion potential prediction (coefficient of determination <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(R^{2}\)</EquationSource> <EquationSource Format="MATHML"><math> <msup> <mi>R</mi> <mn>2</mn> </msup> </math></EquationSource> </InlineEquation> of 0.98 for training and 0.93 for testing), while the Extreme Gradient Boosting model excelled in predicting corrosion current density (coefficient of determination <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(R^{2}\)</EquationSource> <EquationSource Format="MATHML"><math> <msup> <mi>R</mi> <mn>2</mn> </msup> </math></EquationSource> </InlineEquation> of 0.97 for training and 0.94 for testing). Notably, validation through independent electrochemical experiments demonstrated the models’ excellent generalization ability, with prediction errors for corrosion potential and current density within 2% and 6.6%, respectively. Furthermore, Shapley Additive Explanations analysis identified Nd, Y, and Ca as key alloying elements, while&#xa0;Cl<sup>−</sup> concentration and temperature were revealed as core environmental drivers. Finally, a Python-based graphical user interface was developed to provide an intuitive and rapid corrosion prediction tool for engineering applications.</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

Machine learning with local outlier factor for predicting the electrochemical corrosion behavior of rare-earth-doped Mg alloys in chloride ions

  • 期刊论文

  • 发表日期：

    2026年1月8日

查看全文
  • Qiao Huang,
  • Yucheng Cao,
  • Atwakyire Moses,
  • Hao Wang,
  • Wenhao Li,
  • Ce Jiao,
  • Ding Chen

摘要

Rare earth-containing magnesium alloys are critical materials in biomedical applications, yet their corrosion performance directly determines service safety. To overcome the time-consuming limitations of traditional experiments and the difficulty in quantifying complex corrosion mechanisms, this study established a machine learning prediction framework using literature-derived alloy compositions and environmental data. Six algorithms, including Random Forest Regressor, Extreme Gradient Boosting, and Support Vector Machine, were rigorously evaluated. Beyond standard grid search, an advanced optimization strategy integrating the Local Outlier Factor method for noise reduction and learning curve analysis was employed to effectively mitigate overfitting. The results indicate that the optimized Random Forest Regressor model achieved the highest accuracy for corrosion potential prediction (coefficient of determination \(R^{2}\) R 2 of 0.98 for training and 0.93 for testing), while the Extreme Gradient Boosting model excelled in predicting corrosion current density (coefficient of determination \(R^{2}\) R 2 of 0.97 for training and 0.94 for testing). Notably, validation through independent electrochemical experiments demonstrated the models’ excellent generalization ability, with prediction errors for corrosion potential and current density within 2% and 6.6%, respectively. Furthermore, Shapley Additive Explanations analysis identified Nd, Y, and Ca as key alloying elements, while Cl− concentration and temperature were revealed as core environmental drivers. Finally, a Python-based graphical user interface was developed to provide an intuitive and rapid corrosion prediction tool for engineering applications.