An intelligent ratiometric fluorescence platform assisted by machine learning for sensitive detection of Cr(VI)

Cr(VI), a highly toxic and carcinogenic contaminant, presents significant risks to ecological and human health due to its pervasive presence in food and environmental systems. Even minimal exposure can induce oxidative stress and genotoxicity, underscoring the urgent need for rapid, sensitive, and field-deployable analytical methods. In this study, we developed a ratiometric fluorescence probe based on a blue-emitting metal-organic framework (MOF) as the reference signal and a red-emitting CdTe quantum dot (QDs) as the responsive element. Upon Cr(VI) exposure, red fluorescence is quenched through inner filter effect (IFE) and electrostatic interaction (EI), while blue fluorescence remains unaltered, resulting in a visible color change from pink to blue. To achieve precise quantification, the ratiometric system was integrated with three representative ML algorithms—Random Forest (RF), Support Vector Regression (SVR), and BP Neural Network (BPNN). Smartphone-captured fluorescence images were processed by extracting RGB values to train and validate the ML models. Among them, the RF algorithm demonstrated superior performance, achieving an R 2 value of 0.9942, root mean square error of prediction of 3.5842, mean absolute error of prediction of 2.1893, and 93.31 % prediction accuracy. The proposed method achieved satisfactory recovery rates of 98.81–134.00 % and relative standard deviations below 4.32 % in real food and environmental samples. This study presents an intelligent, dual-mode strategy integrating ratiometric fluorescence sensing with machine learning for trace Cr(VI) detection. The approach offers high sensitivity, accuracy, and visual recognition, providing a generalizable platform for automated, on-site, and visualized analysis in complex food and environmental matrices.