The specific recognition of glucose in real samples by AC-impedance method assisted with 3-aminophenylboronic acid-modified magnetic nanoparticles and predictive analysis via GridSearch-XGBoost

An electrochemical impedance detection method based on 3-aminophenylboronic acid-modified magnetic nanoparticles (M-APBA) is proposed, with machine learning models employed to facilitate high-sensitivity and high-accuracy glucose detection in complex biological samples. By leveraging the specific binding affinity between synthesized M-APBA nanoparticles and glucose, detection sensitivity and anti-interference capability were significantly enhanced. Experimental results demonstrated a well-defined linear response for glucose concentrations ranging from 1 to 50 μmol L⁻¹, with a detection limit as low as 0.71 μmol L⁻¹. Multidimensional impedance spectral features were extracted using the Randles equivalent circuit model, and the prediction accuracy of glucose concentration in real sample was significantly improved by incorporating an XGBoost model optimized via grid search method. The model achieved an R² value of 0.9997, with mean absolute error (MAE) and root mean square error (RMSE) reduced to 0.1068 and 0.2321, respectively. Compared with traditional single-feature fitting methods (the Rct value), the proposed multi-feature fusion approach demonstrated superior detection accuracy and stability in complex biological samples. Furthermore, measurements conducted on real samples revealed that M-APBA nanoparticle-modified electrodes maintained excellent detection performance in complex environments, exhibiting high repeatability and strong anti-interference capability. This study proposes a non-enzymatic glucose detection strategy by integrating M-APBA-modified magnetic nanoparticles with a machine learning-based impedance analysis framework, offering a novel solution for non-invasive, real-time glucose monitoring and demonstrating great potential for application in portable, low-cost wearable detection devices.