Machine learning-guided synthesis of nitrogen-doped carbon dots for fluorometric determination of Fe(III) in water

A machine learning (ML)-assisted strategy has been applied to optimize nitrogen-doped carbon dots  (N-CDs) synthesized via a solvothermal route using glutathione, urea, and formamide as the specific precursor system. A dataset of 250 samples was constructed through systematic experimentation and data augmentation to train XGBoost, Random Forest, and LightGBM models for predicting fluorescence intensity. Among these models, XGBoost showed the best predictive performance with an R² value of 0.9032. Model analysis suggested that urea dosage, the GSH-to-urea ratio, and reaction temperature were the most influential factors within the investigated parameter space. Guided by the ML-assisted optimization, the obtained N-CDs had an average particle size of 3.64 nm, a quantum yield of 8.99%, and abundant surface functional groups. The optimized N-CDs enabled Fe³⁺ detection within 1 min, likely through a synergistic static-dynamic quenching process, with a limit of detection of 0.89 µM and a linear range of 5-35 µM. In real water samples, the recoveries ranged from 95.68% to 104.10%, with RSD values below 3.85%, indicating good applicability in the tested matrices. Furthermore, N-CDs were incorporated into agarose hydrogels to prepare a portable visual sensor for Fe(III) detection over the range 0.5-10 µM. This work demonstrates the usefulness of an ML-assisted approach for optimizing this specific N-CD synthesis system and its application in the detection of Fe(III).