基于KNN算法建立晒后皮肤状态评估模型

doi:10.3969/j.issn.2097-2806.2025.03.011

摘要/Abstract

摘要：

探索不同剂量紫外线照射后皮肤指标变化趋势，建立晒后皮肤状态评估模型。首先，筛选出变化有规律且具灵敏性的指标，优化黑化模型进一步扩大样本库，利用临床专家对晒后皮肤状态的分级作为学习标准，基于K邻近分类算法（KNN）对指标数据进行训练识别，建立晒后皮肤状态分级评估模型，经10折交叉验证后超参数K=3时，模型的mmce均值为0.015，预测精度acc均值为0.985，预测的准确度高达98.5%。结果表明，该模型能够将晒后皮肤状态的主观评级客观量化，高效率、高精度识别晒后皮肤状态。研究结果可为晒后皮肤状态评估和晒后修护功效评价体系提供技术支持。

关键词: 日晒, 皮肤状态, 黑化模型, KNN算法

Abstract:

This study established a model for assessing post-sunburn skin condition by exploring the trends of skin indexes after different doses of UV irradiation. First, we screened out the indicators with regular and sensitive changes, optimized the tanning model to further expand the sample library, used the clinical experts’ grading of post-tanning skin status as the learning standard, and trained the identification of the indicator data based on the K Neighborhood Classification Algorithm （KNN） to establish the post-tanning skin status grading assessment model. After 10-fold cross validation, when the hyperparameter K=3, the mmce mean value of the model is 0.015, and the mean value of the prediction accuracy acc is 0.985, with a prediction accuracy of 98.5%. The results show that the model is able to objectively quantify the subjective ratings of post-tanning skin condition and recognize post-tanning skin condition with high efficiency and accuracy. The results can provide technical support for the assessment of post-sunburn skin condition and the evaluation system of post-sunburn repair efficacy.

Key words: sun exposure, skin condition, tanning model, KNN algorithm

中图分类号:

TQ658

李以洪, 许梦然, 盘瑶, 吴金昊, 刘琦, 常思思, 赵华. 基于KNN算法建立晒后皮肤状态评估模型[J]. 日用化学工业（中英文）, 2025, 55(3): 349-357.

Yihong Li, Xu Mengran, Pan Yao, Wu Jinhao, Liu Qi, Chang Sisi, Zhao Hua. Modeling of post-sunburn skin condition assessment based on KNN algorithm[J]. China Surfactant Detergent & Cosmetics, 2025, 55(3): 349-357.

图/表 21

表1

表2

表3

表4

表5

表6

表7

表8

图1

图2

图3

图4

图5

图6

图7

图8

图9

表9

图10

图11

图12

参考文献 28

[1]	Elżbieta L, Joanna B, Beata M. Occupational exposure to natural UV radiation and premature skin ageing[J]. International Journal of Occupational Safety and Ergonomics: Jose, 2014, 20 (4): 639-645.
[2]	Yang Jingwen, Fan Guobiao, Tan Fei, et al. The role and safety of UVA and UVB in UV-induced skin erythema[J]. Frontiers in Medicine, 2023, 10: 1163697.
[3]	Young R A. Acute effects of UVR on human eyes and skin[J]. Progress in Biophysics and Molecular Biology, 2006, 92 (1): 80-85.
[4]	Lopes D M, Mcmahon S B. Ultraviolet radiation on the skin: a painful experience?[J]. CNS Neuroscience & Therapeutics, 2016, 22 (2): 118-126.
[5]	Tang Xiaoyou, Yang Tianyi, Yu Daojiang, et al. Current insights and future perspectives of ultraviolet radiation (UV) exposure: Friends and foes to the skin and beyond the skin[J]. Environment International, 2024, 185: 108535.
[6]	Merin K, Shaji M, Kameswaran R. A review on sun exposure and skin diseases[J]. Indian Journal of Dermatology, 2022, 67 (5): 625.
[7]	Bustamante M, Hernandez F C, Tewari A, et al. Dose and time effects of solar-simulated ultraviolet radiation on the in vivo human skin transcriptome[J]. The British Journal of Dermatology, 2020, 182 (6): 1458-1468.
[8]	Narayanan D L, Saladi R N, Fox J L. Ultraviolet radiation and skin cancer[J]. International Journal of Dermatology, 2010, 49 (9): 978-986. doi: 10.1111/j.1365-4632.2010.04474.x pmid: 20883261
[9]	于笑乾, 贾焱, 李钧翔, 等. 生物信息学及数字技术在化妆品领域的应用概况[J]. 日用化学工业(中英文), 2022, 52 (12): 1359-1365.
[10]	李昌, 王鑫, 冯周, 等. 基于KNN分类算法的微生物白云岩岩相测井综合识别——以四川盆地GM地区灯四段为例[J]. 海相油气地质, 2023, 28 (4): 433-440.
[11]	隋文涛, 王文超, 袁林, 等. 基于KNN算法的铣刀状态监测技术研究[J]. 机械设计与制造, 2023 (6): 89-91, 96.
[12]	刘宏超, 马俊杰, 李韧. 基于KNN机器学习方法对青藏高原唐古拉地区表层土壤水热状况的模拟[J]. 冰川冻土, 2021, 43 (4): 1243-1252. doi: 10.7522/j.issn.1000-0240.2021.0079
[13]	尹月煊, 赵华. 化妆品功效评价(Ⅰ)——化妆品功效宣称的科学支持[J]. 日用化学工业, 2018, 48 (1): 8-13.
[14]	盘瑶, 赵华. 化妆品功效评价(Ⅵ)——化妆品人体功效评价的实验设计[J]. 日用化学工业, 2018, 48 (6): 314-321.
[15]	赵华, 王楠. 化妆品功效评价(Ⅲ)——美白功效宣称的科学支持[J]. 日用化学工业, 2018, 48 (3): 129-133, 139.
[16]	Lin Yichun, Chen Yujie, Huang Bingfang, et al. Acute dermal effects of solar UV irradiation and efficacy of sunscreen use[J]. Environmental Pollutants and Bioavailability, 2022, 34 (1): 456-468.
[17]	刘玮. 皮肤日晒红斑的光生物学基础[J]. 中华医学信息导报, 2005 (4): 21.
[18]	刘玮. 皮肤日晒红斑[J]. 临床皮肤科杂志, 2005 (7): 483-484.
[19]	Chun Y L, Chieh Y C, Fang B H, et al. Acute dermal effects of solar UV irradiation and efficacy of sunscreen use[J]. Environmental Pollutants and Bioavailability, 2022, 34 (1): 456-468.
[20]	Langman L. Portable erythema meter and its application to use in human skin[J]. Int J Cosmet, 2010, 12 (2): 63-70.
[21]	Claire B, Michèle V. Cutaneous solar ultraviolet exposure and clinical aspects of photodamage[J]. Indian Journal of Dermatology, Venereology and Leprology, 2012, 78 (7): S9-S14.
[22]	Dong Qiang, Ma Xue, Xian Zhiming, et al. Efficacy evaluation of a cream containing fresh milk to alleviate skin erythema reaction[J]. Daily Chemical Science, 2021, 44 (2): 22-25, 32.
[23]	董强, 马雪, 羡志明, 等. 一款含鲜奶面霜缓解皮肤红斑反应的功效评价研究[J]. 日用化学品科学, 2021, 44 (2): 22-25, 32.
[24]	窦小凡. KNN算法综述[J]. 通讯世界, 2018 (10): 273-274.
[25]	Bingjie L, Bo S, Chenchen N, et al. Quantitative timing strategy model based on improved knn algorithm[J]. Procedia Computer Science, 2022, 202: 61-66.
[26]	陈文静, 许诺, 教召航, 等. 机器学习在乳腺癌荧光光谱诊断中的应用研究[J]. 光谱学与光谱分析, 2023, 43 (8): 2407-2412.
[27]	梁子超, 李智炜, 赖铿, 等. 10折交叉验证用于预测模型泛化能力评价及其R软件实现[J]. 中国医院统计, 2020, 27 (4): 289-292.
[28]	李刚, 高武奇, 杨瑞臣. 有指导机器学习超参数的交叉验证智能优化[J]. 西安工业大学学报, 2016, 36 (11): 906-910.

表现	等级
不均匀轻微红斑	1
均匀轻微红斑	2
中度红斑	3
深色重度红斑	4

测试时间/剂量	TEWL值/（g/（h·m²））
测试时间/剂量	0	1.5 倍MED	2 倍MED	2.5 倍MED
建模前	8.61±1.67	9.05±2.70	8.84±2.11	8.90±1.72
即时	8.74±1.65	8.71±1.69	8.54±1.71	8.99±1.86
2 h	9.94±2.16	9.57±1.75	9.57±1.90	9.49±1.84
4 h	9.79±2.60	8.85±2.02	8.90±2.06	9.25±1.92
6 h	9.75±1.72	9.27±1.70	9.42±2.07	9.79±2.19
24 h	9.63±2.14	9.65±1.87	9.66±1.88	9.69±1.75
d3	8.91±2.26	7.97±1.69	8.38±1.96	8.60±2.51
d7	9.15±2.40	8.88±3.00	9.45±3.56	9.06±2.91
d14	9.51±1.66	9.07±1.68^d	9.37±1.79	10.10±2.10^b
d28	9.03±1.73	8.47±2.04	8.40±1.92	8.80±1.94

测试时间/剂量	SCH值/a.u.
测试时间/剂量	0	1.5 倍MED	2 倍MED	2.5 倍MED
建模前	54.95±11.85	58.22±11.94	55.78±11.16	53.73±12.09
即时	56.73±10.01	58.95±10.86	55.83±8.26	54.85±7.83
2 h	59.33±9.31	59.90±8.33^d	57.60±8.54	55.73±7.94^b
4 h	57.83±11.01	57.78±10.69	56.07±9.45	55.88±8.64
6 h	58.48±11.63	60.08±9.31	59.68±10.03	59.80±8.73
24 h	60.05±8.32	60.93±8.72	57.98±8.90	57.83±8.62
d3	57.90±9.98	57.40±10.16	54.93±9.02	55.05±9.33
d7	60.10±8.73^cd	56.80±10.99^cd	51.13±10.24^abd	45.23±9.42^abc
d14	53.35±10.40^cd	49.78±9.78^d	47.30±12.10^a	43.48±11.19^ab
d28	53.67±10.16	52.73±11.38	53.63±9.91	54.02±10.57

测试时间/剂量	EI值
测试时间/剂量	0	1.5 倍MED	2 倍MED	2.5 倍MED
建模前	149.23±49.08	150.10±47.53	141.38±54.01	133.23±38.15
即时	165.46±38.65	158.24±43.32	166.39±40.28	165.89±47.42
2 h	170.59±38.48^cd	184.86±47.59^cd	211.15±50.35^ab	219.14±53.99^ab
4 h	162.22±36.31^bcd	220.29±60.97^acd	266.49±60.50^abd	293.70±49.06^abc
6 h	169.66±36.69^bcd	259.13±56.04^acd	302.34±59.47^ab	323.08±59.49^ab
24 h	157.36±66.93^bcd	273.22±44.83^acd	338.88±71.68^abd	372.90±47.28^abc
d3	144.13±46.85^bcd	266.24±49.41^acd	333.81±57.43^abd	378.32±57.23^abc
d7	152.28±35.87^bcd	224.89±51.43^acd	305.35±54.65^abd	352.09±54.87^abc
d14	140.12±40.83^bcd	209.55±51.48^acd	265.22±47.44^abd	301.32±40.43^abc
d28	145.23±30.90^bcd	201.84±40.90^acd	241.22±42.20^abd	273.54±49.86^abc

测试时间/剂量	MI值
测试时间/剂量	0	1.5 倍MED	2 倍MED	2.5 倍MED
建模前	77.62±37.51	77.00±33.50	83.03±33.30	75.54±30.07
即时	91.20±31.07	91.71±31.76	94.08±32.08	87.85±32.46
2 h	86.26±37.69	90.75±40.01	93.67±32.84	80.48±36.45
4 h	92.57±36.94	90.34±43.07	92.28±43.60	80.48±38.95
6 h	91.25±37.52	84.03±39.95	87.33±39.45	79.41±37.62
24 h	88.83±36.33^d	79.62±43.35	72.52±45.70	65.14±38.85^a
d3	84.34±28.52	91.08±37.41	87.55±42.74	73.62±37.59
d7	80.13±31.13^bcd	120.08±27.08^acd	149.87±39.65^ab	168.86±50.82^ab
d14	78.39±30.44^bcd	119.40±26.99^acd	153.25±30.42^abd	168.74±29.91^abc
d28	79.54±31.58^bcd	119.58±34.28^acd	143.35±35.11^ab	158.01±33.76^ab