Quantitative Analysis of Diabetic Blood Raman Spectroscopy Based on XGBoost

Ming-xuan WANG; Qiao-yun WANG; Fei-fei PIAN; Peng SHAN; Zhi-gang LI; Zhen-he MA

doi:10.3964/j.issn.1000-0593(2022)06-1721-07

Journals >Spectroscopy and Spectral Analysis >Volume 42 >Issue 6 >Page 1721 > Article

Spectroscopy and Spectral Analysis
Vol. 42, Issue 6, 1721 (2022)

Quantitative Analysis of Diabetic Blood Raman Spectroscopy Based on XGBoost

Ming-xuan WANG^*, Qiao-yun WANG^*;, Fei-fei PIAN, Peng SHAN, Zhi-gang LI, and Zhen-he MA

Author Affiliations

College of Information Science and Engineering, Northeastern University, Shenyang 110819, China

show less

DOI: 10.3964/j.issn.1000-0593(2022)06-1721-07 Cite this Article

Ming-xuan WANG, Qiao-yun WANG, Fei-fei PIAN, Peng SHAN, Zhi-gang LI, Zhen-he MA. Quantitative Analysis of Diabetic Blood Raman Spectroscopy Based on XGBoost[J]. Spectroscopy and Spectral Analysis, 2022, 42(6): 1721 Copy Citation Text

show less

Fig. 1. Original Raman spectra of blood samples

Download full size | View in the Article

Fig. 2. Schematic diagram of XGBoost algorithm

Download full size | View in the Article

Fig. 3. XGBoost Hyper-parameters tuning

Download full size | View in the Article

Fig. 4. Model prediction regression diagram
(a): DT model; (b): SVR model; (c): XGBoost model; (d): RF model

Download full size | View in the Article

Fig. 5. Clarke grid error analysis diagram
(a): DT Clark grid error analysis; (b): SVR Clark grid error analysis; (c): XGBoost Clark grid error analysis; (d): RF Clark grid error analysis

Download full size | View in the Article

Algorithm: Exact Greedy Algorithm for Split Finding

Input: I, instance set of current node
Input: d, feauure dimension

\begin{array}{l} gain \leftarrow 0 \\ G \leftarrow \sum_{i \in I} g_{i}, H \leftarrow \sum_{i} h_{i} \end{array}

for k=1 to m do
G_L←0, H_L←0
for j in sorted (I, by x_jk) do
G_L←G_L+g_j, H_L←H_L+h_j
G_R←G-G_L, H_R←H-H_L
score←max

(score, \frac{G_{L}^{2}}{H_{L} + λ} + \frac{G_{R}^{2}}{H_{R} + λ} - \frac{G^{2}}{H + λ})

end
Output: Split with max score

Table 1. Schematic diagram of greedy algorithm

View in the Article

超参数	超参数含义	数值
learning_rate	在迭代更新中使用步长收缩防止过度拟合	0.3
n_estimators	模型中增加树的数量	150
max_depth	树的最大深度	7
min_child_weight	在树构建过程中进一步划分叶节点	5
subsample	每棵树对样本的采样率	1
colsample_bytree	列采样率	1
gamma	损失函数比较值	0
reg_alpha	L1正则化项权重	0
reg_lambda	L2正则化权重项	1

Table 2. Hyper-parameters

View in the Article

模型	R²	RMSEC	RMSEP	RPD
RF	0.873 01	0.841 34	0.981 95	2.370 43
SVR	0.821 99	0.953 56	1.067 79	2.805 87
XGBoost	0.999 99	0.007 49	0.007 17	331.973 18
DT	0.908 57	0.960 41	0.878 49	3.308

Table 3. Model evaluation

Download Citation

Save the article for my favorites

Paper Information