Fluorescent sensing of amoxicillin in milk using carbon quantum dots: optimization by design of experiments and machine learning

Carbon quantum dots (CQDs) synthesized via a approach were comprehensively characterized using advanced techniques including FTIR, TEM, SEM, EDX, and fluorescence spectroscopy, confirming their nanoscale structure and functional surface groups. Experimental design was optimized using response surface methodology (RSM) and central composite design (CCD), where pH, reaction time, temperature, and sensor dosage were systematically evaluated to maximize the fluorescence quenching response (F₀/F). Statistical analysis revealed excellent model adequacy (R² = 0.9834, Adj-R² = 0.9680, Pred-R² = 0.9082), with pH, reaction time, and sensor dosage identified as the most influential parameters. Diagnostic plots verified the assumptions of linearity, normality, and homoscedasticity, ensuring predictive reliability. To further validate and enhance predictive performance, machine learning models were employed. Random Forest demonstrated superior accuracy (R² ≈ 0.96) and lower error metrics compared to quadratic regression, highlighting the potential of ensemble learning in capturing nonlinear multivariate interactions. Calibration studies established two linear ranges (10–300 µM and 300–650 µM), with detection limits (LOD) of 41.47 µM and 101.78 µM and quantification limits (LOQ) of 125.66 µM and 308.43 µM, respectively. Blank precision was satisfactory (RSD = 3.39%), confirming reproducibility. Interference studies showed negligible effects from common ions and biomolecules, validating the sensor’s selectivity. So, the integration of structural characterization, DOE-based optimization, and machine learning evaluation underscores the robustness of the CQDs-based sensor as a sensitive, selective, platform for amoxicillin detection in complex matrices.