Drugs Identification Using Near-Infrared Spectroscopy Based on Random Forest and CatBoost

Ping JIANG; Hao-xiang LU; Zhen-bing LIU

doi:10.3964/j.issn.1000-0593(2022)07-2148-08

Spectroscopy and Spectral Analysis, Volume. 42, Issue 7, 2148(2022)

Drugs Identification Using Near-Infrared Spectroscopy Based on Random Forest and CatBoost

Ping JIANG1,1;... Hao-xiang LU2,2; and Zhen-bing LIU2,2; *; |Show fewer author(s)

Author Affiliations

¹1. School of Computer and Information Technology, Guangxi Police College, Nanning 530028, China

²2. College of Computer and Information Security, Guilin University of Electronic Technology, Guilin 541004, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(9)

Fig. 1. The structure of RF-CatBoost

Download full size

View in Article

Fig. 2. NIR spectra of pretreated cefixime tablets

Download full size

View in Article

Fig. 3. Covariance matrix of drug NIR data before (a) and after (b) pretreatment

Download full size

View in Article

Fig. 4. Classification accuracy of different decision tree numbers in Catboost on datasets of different sizes in group A (a) and group B (b)

Download full size

View in Article

Fig. 5. Standard deviations of each model on different sizes data sets in group A (a) and group B (b)

Download full size

View in Article

Table 1. Near infrared spectral data of cefixime tablets

View table

View in Article

Table 1. Near infrared spectral data of cefixime tablets

厂商	非铝塑包装	铝塑包装	合计
湖南方盛制药股份有限公司	54	54	108
江苏正大清江制药有限公司	63	56	119
山东鲁抗医药股份有限公司	51	40	91
山东罗欣药业股份有限公司	48	48	96
共计	216	198	414

Table 2. Configuration table of different number of training sets in group A and B
View table
View in Article
Table 2. Configuration table of different number of training sets in group A and B
数据集样本总数正样本数负样本数
A 40 15 25
60 20 40
80 25 55
100 30 70
120 35 85
140 40 100
160 45 115
180 50 130
B 30 10 20
50 15 35
70 20 50
90 25 65
110 30 80
130 35 95
150 40 110
170 45 125

Table 3. Classification accuracy of each model on different sizes data sets in group A and B (%)

View table

View in Article

Table 3. Classification accuracy of each model on different sizes data sets in group A and B (%)

组别	训练/测试集	ELM	SWELM	SVM	BP	Boosting	CatBoost	CatBoost RF-CatBoost
A	40/176	92.36	92.33	93.68	89.36	94.44	94.99	96.79
	60/156	93.65	93.88	94.09	90.31	94.95	95.03	97.89
	80/136	94.99	95.11	95.35	91.01	96.22	96.85	98.82
	100/116	96.03	96.29	96.88	91.85	97.05	97.52	99.05
	120/96	97.88	97.95	97.23	92.99	97.99	97.89	99.95
	140/76	97.99	97.64	98.88	93.35	98.85	98.98	100
	160/56	98.05	98.01	99.05	94.99	99.01	99.02	100
	180/36	98.88	98.91	99.01	95.89	99.18	99.35	100
B	30/168	91.28	90.99	92.34	88.75	92.86	93.95	95.97
	50/148	92.69	91.67	93.38	90.31	94.01	94.98	96.59
	70/128	93.19	93.11	94.20	91.11	95.19	96.82	98.79
	90/108	94.25	94.26	95.38	91.85	96.21	96.98	99.92
	110/88	94.95	95.89	96.21	92.99	96.89	97.99	100
	130/68	95.92	97.22	98.09	93.35	97.96	98.09	100
	150/48	97.95	98.09	98.89	94.99	98.39	98.19	100
	170/28	98.88	98.85	99.00	95.89	99.08	98.51	100

Table 4. Runningtime of each model on different sizes data sets in group A and group B (s)

View table

View in Article

Table 4. Runningtime of each model on different sizes data sets in group A and group B (s)

组别	训练/测试集	ELM	SWELM	SVM	BP	Boosting	CatBoost	CatBoost RF-CatBoost
A	40/176	0.009 4	0.003 1	0.017 0	38.098 8	15.928 8	8.717 0	6.088 3
	60/156	0.013 0	0.005 0	0.032 1	38.339 7	17.985 4	15.903 0	7.408 3
	80/136	0.013 8	0.015 8	0.063 3	38.449 8	20.111 6	23.147 4	9.237 4
	100/116	0.021 1	0.017 1	0.113 8	38.864 7	22.273 0	30.441 0	9.365 9
	120/96	0.031 1	0.030 2	0.151 8	39.800 8	24.884 6	38.090 2	10.297 3
	140/76	0.044 2	0.040 1	0.219 8	40.885 0	26.690 6	45.062 6	10.988 8
	160/56	0.056 7	0.059 7	0.287 8	41.297 8	28.565 2	52.382 0	12.083 4
	180/36	0.078 7	0.074 0	0.359 5	42.380 7	30.661 2	59.526 4	12.530 8
B	30/168	0.017 5	0.009 4	0.007 3	2.171 0	2.510 6	1.496 2	0.477 1
	50/148	0.053 5	0.023 7	0.024 3	4.159 9	3.322 7	2.395 7	1.164 1
	70/128	0.127 5	0.053 3	0.052 3	6.208 0	4.081 7	3.325 8	2.031 3
	90/108	0.215 5	0.098 3	0.102 7	8.350 6	5.336 2	4.422 0	2.907 4
	110/88	0.339 1	0.167 0	0.177 8	10.416 0	6.174 3	5.406 6	3.831 2
	130/68	0.495 8	0.260 2	0.295 3	12.404 0	6.887 9	6.411 8	4.785 0
	150/48	0.706 1	0.406 3	0.439 3	14.357 2	7.694 5	7.363 1	5.779 7
	170/28	0.912 0	0.616 8	0.643 0	16.322 8	8.565 5	8.346 4	6.829 1

Tools

Get Citation

Copy Citation Text

Ping JIANG, Hao-xiang LU, Zhen-bing LIU. Drugs Identification Using Near-Infrared Spectroscopy Based on Random Forest and CatBoost[J]. Spectroscopy and Spectral Analysis, 2022, 42(7): 2148

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Orginal Article

Received: Jan. 12, 2022

Accepted: --

Published Online: Nov. 16, 2022

The Author Email:

DOI:10.3964/j.issn.1000-0593(2022)07-2148-08

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Near infrared spectral data of cefixime tablets

Table 1. Near infrared spectral data of cefixime tablets

Table 2. Configuration table of different number of training sets in group A and B

Table 2. Configuration table of different number of training sets in group A and B

Table 3. Classification accuracy of each model on different sizes data sets in group A and B (%)

Table 3. Classification accuracy of each model on different sizes data sets in group A and B (%)

Table 4. Runningtime of each model on different sizes data sets in group A and group B (s)

Table 4. Runningtime of each model on different sizes data sets in group A and group B (s)

微信扫一扫：分享