Emerging applications of fluorescence excitation-emission matrix with machine learning for water quality monitoring: A systematic review

Fluorescence excitation-emission matrix (FEEM) spectroscopy is increasingly utilized in water quality monitoring due to its rapid, sensitive, and non-destructive measurement capabilities. The integration of machine learning (ML) techniques with FEEM offers a powerful approach to enhance data interpretation and improve monitoring efficiency. This review systematically examines the application of ML-FEEM in urban water systems across three primary tasks of ML: classification, regression, and pattern recognition. Contributed by the effectiveness of ML in nonlinear and high dimensional data analysis, ML-FEEM achieved superior accuracy and efficiency in pollutant qualification and quantification. The fluorescence features extracted through ML are more representative and hold potential for generating new FEEM samples. Additionally, the rich visualization capabilities of ML-FEEM facilitate the exploration of the migration and transformation of dissolved organic matter in water. This review underscores the importance of leveraging the latest ML advancements to uncover hidden information within FEEM data, and advocates for the use of pattern recognition methods, represented by self-organizing map, to further elucidate the behavior of pollutants in aquatic environments. Despite notable advancements, several issues require careful consideration, including the portable or online setups for FEEM collection, the standardized pretreatment processes for FEEM analysis, and the smart feedback of long-term FEEM governance.