Scholarlink

<p>Accurate prediction of both crystallite size (D) and band gap energy (Eg) is crucial for tailoring the functional properties of anatase TiO<sub>2</sub>. In this study, we evaluate the predictive performance of Linear Regression (LR), Random Forest (RF), and Gaussian Process Regression (GPR) using synthesis temperature and lattice constants (a, c) as input features. Statistical evaluations reveal that GPR consistently yields the highest accuracy for both targets, achieving R<sup>2</sup> values of 0.9790 for crystallite size and 0.9996 for band gap energy. The LR model showed significant limitations, particularly in predicting Eg (R<sup>2</sup> = 0.1120) and D (R<sup>2</sup> = 0.5313), due to strong multicollinearity between lattice parameters (<i>r</i> = -0.99) and the inherently nonlinear nature of crystal growth and electronic modulation. While RF demonstrated robust performance (R<sup>2</sup> &gt; 0.90 for both targets), it remained less precise than the GPR framework. These findings highlight GPR as a superior computational tool for modeling complex, nonlinear diffusion-strain regimes in semiconductor materials, thereby enabling accelerated characterization of TiO<sub>2</sub> nanostructures.</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-driven modeling of crystallite growth and optical band gap in anatase TiO2

  • 期刊论文

  • 发表日期：

    2026年5月20日

查看全文
  • Vicran Zharvan

摘要

Accurate prediction of both crystallite size (D) and band gap energy (Eg) is crucial for tailoring the functional properties of anatase TiO2. In this study, we evaluate the predictive performance of Linear Regression (LR), Random Forest (RF), and Gaussian Process Regression (GPR) using synthesis temperature and lattice constants (a, c) as input features. Statistical evaluations reveal that GPR consistently yields the highest accuracy for both targets, achieving R2 values of 0.9790 for crystallite size and 0.9996 for band gap energy. The LR model showed significant limitations, particularly in predicting Eg (R2 = 0.1120) and D (R2 = 0.5313), due to strong multicollinearity between lattice parameters (r = -0.99) and the inherently nonlinear nature of crystal growth and electronic modulation. While RF demonstrated robust performance (R2 > 0.90 for both targets), it remained less precise than the GPR framework. These findings highlight GPR as a superior computational tool for modeling complex, nonlinear diffusion-strain regimes in semiconductor materials, thereby enabling accelerated characterization of TiO2 nanostructures.