An integrated PCA-XGBoost method for fluorescence spectral analysis of marine microplastics

Microplastic identification based on fluorescence spectroscopy remains challenging because of spectral overlap and spectral variations associated with aging, which can affect classification performance. To address these challenges, an interpretable PCA-XGBoost framework is proposed for fluorescence spectral analysis, integrating principal component analysis (PCA) with XGBoost classification. By linking PCA loading profiles with XGBoost feature importance, the framework enables the identification of characteristic wavelengths associated with microplastic discrimination and facilitates interpretation of the spectral features underlying classification decisions. The method was evaluated using both pure and mixed microplastic samples representing multiple polymer categories under aged and unaged conditions. The proposed framework achieved an overall classification accuracy of 99.94%, while exhibiting stable feature-importance patterns across the evaluated analyses. Furthermore, the PCA-XGBoost model developed using laboratory microplastic samples was applied to real marine microplastics, demonstrating its potential for practical environmental applications.