Scholarlink

<p>This study evaluates the performance of two machine learning models—Random Forest (RF) and Artificial Neural Network (ANN)—for predicting tensile strength based on four input parameters: B<sub>4</sub>C Composition, Compaction Pressure, Sintering Temperature, and Sintering Time. The models were optimized and assessed using regression metrics, residual diagnostics, and feature importance analysis. The Random Forest model demonstrated superior predictive accuracy, achieving an R² value of 0.9599, Mean Absolute Error (MAE) of 8.5781, and Root Mean Squared Error (RMSE) of 10.1620, compared to the ANN model. Residual diagnostics revealed that the Random Forest model had more consistently distributed residuals and fewer large errors, aligning with its focus on Compaction Pressure and B₄C Composition as the most influential features. In contrast, the ANN model showed a wider spread of residuals and less emphasis on B<sub>4</sub>C Composition, which contributed to its comparatively lower performance. Microstructural analysis supported the significance of Compaction Pressure in enhancing tensile strength and validated the Random Forest model’s feature prioritization. The results indicate that Random Forest is more robust and reliable for this prediction task, while the ANN model, though capable, requires further optimization to improve accuracy. This study highlights the importance of selecting appropriate models and feature sets for accurate material property predictions.</p> Graphical abstract on superabsorbent polymers in cementitious materials <p></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 approaches for tensile strength prediction in Al/B4C metal matrix composites: a comparative analysis of neural networks and ensemble methods

  • 期刊论文

  • 发表日期：

    2025年7月2日

查看全文
  • Guttikonda Manohar

摘要

This study evaluates the performance of two machine learning models—Random Forest (RF) and Artificial Neural Network (ANN)—for predicting tensile strength based on four input parameters: B4C Composition, Compaction Pressure, Sintering Temperature, and Sintering Time. The models were optimized and assessed using regression metrics, residual diagnostics, and feature importance analysis. The Random Forest model demonstrated superior predictive accuracy, achieving an R² value of 0.9599, Mean Absolute Error (MAE) of 8.5781, and Root Mean Squared Error (RMSE) of 10.1620, compared to the ANN model. Residual diagnostics revealed that the Random Forest model had more consistently distributed residuals and fewer large errors, aligning with its focus on Compaction Pressure and B₄C Composition as the most influential features. In contrast, the ANN model showed a wider spread of residuals and less emphasis on B4C Composition, which contributed to its comparatively lower performance. Microstructural analysis supported the significance of Compaction Pressure in enhancing tensile strength and validated the Random Forest model’s feature prioritization. The results indicate that Random Forest is more robust and reliable for this prediction task, while the ANN model, though capable, requires further optimization to improve accuracy. This study highlights the importance of selecting appropriate models and feature sets for accurate material property predictions.

Graphical abstract on superabsorbent polymers in cementitious materials