Fluorescent sensor array integrated with machine learning: Robust discrimination of multiple metal ions using carboxyl-rich carbon dots

Abstract

Accurate identification and reliable monitoring of metal ions (MIs) are critically important due to their persistent toxicity, bioaccumulation potential, and significant environmental and public health risks, even at trace concentrations. Herein, we developed a highly sensitive and selective fluorescence sensor array integrated with machine learning. We used a single type of carboxyl-rich carbon dots (CR-dots) combined with multiple buffer environments to achieve rapid, simultaneous discrimination of multiple metal ions (Ag+, Al3+, Cd2+, Co2+, Cr(VI), Cr3+, Cu2+, Fe3+, Fe2+, Hg2+, Mn2+, Ni2+, Pb2+, Sn2+, and Ti4+). We evaluated five machine learning models for classification. These included linear discriminant analysis (LDA), multinomial logistic regression (MLR), support vector machines (SVM), k-nearest neighbors (k-NN) and random forest (RF). Among them, the random forest model achieved the highest performance, with 99.78% accuracy at concentrations as low as 0.05 μM. The array also showed strong classification ability in real water samples, including tap water with added cations and lake water with added anions. Furthermore, the sensor array demonstrated excellent robustness in reliably discriminating complex binary, ternary, and quaternary metal ion mixtures, maintaining high classification accuracy even in challenging real-world environmental samples, including acidic soil leachates and nearshore seawater. This cost-effective, label-free fluorescence sensing platform combines with machine learning algorithms to offer a promising strategy for rapid, high-throughput qualitative screening and monitoring of environmentally relevant metal ions. This highlights substantial potential for practical deployment in environmental protection, pollution assessment, and analytical chemistry.