Machine Learning-Driven Design of Fluorescent Materials: Principles, Methodologies, and Future Directions

Dual-mode fluorescent materials are vital in bioimaging, sensing, displays, and lighting, owing to their efficient emission of visible or near-infrared light. Traditional optimization methods, including empirical experiments and quantum chemical computations, suffer from high costs, high labor intensities, and difficulties capturing complex relationships among molecular structures, synthesis parameters, and key photophysical properties. In this review, fundamental principles, key methodologies, and representative applications of machine learning (ML) in predicting fluorescent material performance are systematically summarized. The core ML techniques covered include supervised regression, neural networks, and physics-informed hybrid frameworks. The representative fluorescent materials analyzed encompass aggregation-induced emission (AIE) luminogens, thermally activated delayed fluorescence (TADF) emitters, quantum dots, carbon dots, perovskites, and inorganic phosphors. This review details the modeling approaches and typical workflows-such as data preprocessing, descriptor selection, and model validation-and highlights algorithmic optimization strategies such as data augmentation, physical constraints embedding, and transfer learning. Finally, prevailing challenges, including limited high-quality data availability, weak model interpretability, and insufficient model transferability, are discussed.