Data on the Construction Processes of Regression Models

This CSV dataset (numbered 1–8) demonstrates the construction processes of the regression models using machine learning methods, which are used to plot Fig. 2–7. The CSV file of 1.LSM_R^2 (plotting Fig. 2) shows the data of the relationship between estimated values and actual values when the least-squares method was used for a model construction. In the CSV file 2.PCR_R^2 (plotting Fig. 3), the number of the principal components was varied from 1 to 5 during the construction of a model using the principal component regression. The data in the CSV file 3.SVR_R^2 (plotting Fig. 4) is the result of the construction using the support vector regression. The hyperparameters were decided by the comprehensive combination from the listed candidates by exploring hyperparameters with maximum R2 values. When a deep neural network was applied to the construction of a regression model, NNeur., NH.L. and NL.T. were varied. The CSV file 4.DNN_HL (plotting Fig. 5a)) shows the changes in the relationship between estimated values and actual values at each  NH.L.. Similarly, changes in the relationships between estimated values and actual values in the case NNeur. or NL.T. were varied in the CSV files 5.DNN_ Neur  (plotting Fig. 5b)) and 6.DNN_LT (plotting Fig. 5c)). The data in the CSV file 7.DNN_R^2 (plotting Fig. 6) is the result using optimal NNeur., NH.L. and NL.T.. In the CSV file 8.R^2 (plotting Fig. 7), the validity of each machine learning method was compared by showing the optimal results for each method.
Experimental conditions
Supply volume of the raw material: 25–125 mL
Addition rate of TiO2: 5.0–15.0 wt%
Operation time: 1–15 min
Rotation speed: 2,200–5,700 min-1
Temperature: 295–319 K
Nomenclature
NNeur.: the number of neurons
NH.L.: the number of hidden layers
NL.T.: the number of learning times

本CSV数据集（编号1至8）展示了利用机器学习方法构建回归模型的过程，并应用于绘制图2至7。其中，编号1的LSM_R^2（绘制图2）CSV文件展示了当采用最小二乘法构建模型时，估算值与实际值之间的关系数据。编号2的PCR_R^2（绘制图3）CSV文件在构建主成分回归模型的过程中，主成分的数量被调整为1至5。编号3的SVR_R^2（绘制图4）CSV文件展示了使用支持向量回归构建模型的结果，其超参数通过从列出的候选者中综合选取，以探索能够实现最大R^2值的超参数。在将深度神经网络应用于回归模型的构建时，神经元数量（NNeur.）、隐藏层数量（NH.L.）和学习次数（NL.T.）被调整。编号4的DNN_HL（绘制图5a）CSV文件展示了在各个NH.L.下估算值与实际值之间关系的变化。编号5的DNN_Neur（绘制图5b）和编号6的DNN_LT（绘制图5c）CSV文件分别展示了在调整NNeur.或NL.T.时估算值与实际值之间关系的变化。编号7的DNN_R^2（绘制图6）CSV文件展示了使用最优的NNeur.、NH.L.和NL.T.所得到的结果。编号8的R^2（绘制图7）CSV文件通过展示每种机器学习方法的最佳结果，比较了各方法的有效性。实验条件如下：原料供应量：25至125毫升；TiO2添加速率：5.0至15.0重量百分比；操作时间：1至15分钟；旋转速度：2,200至5,700转/分钟；温度：295至319开尔文。术语解释：NNeur.：神经元数量；NH.L.：隐藏层数量；NL.T.：学习次数。