3D excitation-emission matrix (EEM) fluorescence-based efficient detection of cosmetics health risks: comparative AI algorithm and eXtreme Gradient Boosting (XGBoost) implementations

The presence of hazardous substances such as heavy metals, toxic organic compounds, and endocrine disruptors in cosmetics poses increasing risks to public health. However, current detection methods like High-Performance Liquid Chromatography (HPLC) and mass spectrometry, while effective, are expensi...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Journal of cleaner production Jg. 533; S. 146996
Hauptverfasser: Du, Runbao, Bu, Te, Huang, Zhenchuan, Morley, John Donald, Xing, Yukun, Li, Meixuan, Xu, Hao, Feng, Yuzhang, Zhou, Xinjie, Long, Dan, Jin, Zhengyu
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Elsevier Ltd 20.11.2025
Schlagworte:
3D fluorescence spectroscopy
Cosmetics
Hazard screening
Machine learning detection
Toxic additives
3D fluorescence spectroscopy
Cosmetics
Toxic additives
Machine learning detection
Hazard screening
ISSN:0959-6526
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:The presence of hazardous substances such as heavy metals, toxic organic compounds, and endocrine disruptors in cosmetics poses increasing risks to public health. However, current detection methods like High-Performance Liquid Chromatography (HPLC) and mass spectrometry, while effective, are expensive, time-consuming, and limited in scope. This study proposes a rapid, low-cost detection framework using three-dimensional fluorescence Excitation-Emission Matrix (EEM) spectroscopy combined with machine and deep learning. A dataset of 1000 EEM spectra covering various pollutants and concentration gradients was used to train and compare four models: Backpropagation Neural Network (BPNN), Convolutional Neural Network (CNN), Random Forest (RF), and eXtreme Gradient Boosting (XGBoost). Among them, XGBoost achieved the highest accuracy (>95%) and robustness in identifying prohibited or excessive additives in both simulated and commercial cosmetic samples. The method significantly reduced detection time from hours to minutes and cut costs by approximately 50% compared to HPLC. A user-friendly interface enabled practical deployment, successfully identifying over 90% of harmful substances in real-world samples. This study demonstrates a scalable, real-time solution for enhancing regulatory monitoring of cosmetic safety. By improving detection speed, accuracy, and accessibility, it contributes meaningfully to consumer protection and public health surveillance.
ISSN:0959-6526
DOI:10.1016/j.jclepro.2025.146996