Machine Learning‐Driven Achieving Efficient Phosphorescent Carbon Nanodots in Aqueous Solution by Suppressing Triplet Electron Leakage

Matrix-assisted synthesis has emerged as a prevalent strategy for the fabrication of solid-state phosphorescent carbon nanodots (CNDs), yet achieving efficient liquid-phase CND systems remains challenging due to complex underlying mechanisms. The knowledge gap poses a substantial barrier to translating phosphorescent CND systems between solid-state and liquid-phase configurations. Herein, the critical role of triplet electron leakage is elucidated in reducing the phosphorescent performance of CNDs and demonstrates how machine learning (ML) can be applied to suppress this leakage, achieving efficient phosphorescent CNDs in aqueous solution. By integrating experimental datasets from hundreds of systematically designed syntheses, an interpretable ML model capable of predicting and tuning phosphorescence lifetimes in CND systems is developed. The quantitative relationships between features and lifetimes using SHapley Additive exPlanations (SHAP) analysis are established, revealing an inverse correlation between matrix thickness and the probability of triplet electron leakage. Guided by the developed ML model, efficient phosphorescent CNDs are achieved in aqueous solution with an emission duration lifetime of exceeding10 s and a phosphorescent quantum yield of over 10%. This study establishes a conceptual and methodological framework for engineering high-performance phosphorescent nanomaterials in liquid-phase systems.