Interpretable machine learning model-driven electrochemical impedance spectroscopic analysis for the determination of...

Rapid advances in materials chemistry and data-driven approaches have accelerated the development of aquatic chemical sensors, yet accurate, real long-term nutrient monitoring remains a significant challenge. Reliable real-time detection of nitrate (NO₃^-), nitrite (NO₂^-), and ammonium (NH₄⁺) ions is essential for understanding aquatic biogeochemistry, mitigating eutrophication, ensuring precision fertigation, and ensuring sustainable water resource and crop management. Conventional electrochemical sensors can achieve low detection limits, but issues of accuracy, reproducibility, and stability under variable conditions hinder their broader application. In this preliminary study, electrochemical impedance spectroscopy (EIS) was employed in a three-electrode system to capture impedance responses over a wide frequency range, where the electrode-electrolyte interface was modelled using an equivalent circuit comprising resistive, capacitive and impedance elements. Impedance features including the real part, imaginary part, amplitude, and phase were analyzed as functions of concentration and frequency for the three target ions. To address the inherent nonlinearities of EIS data, advanced machine learning models were applied, with extreme gradient boosting (XGBoost) used for feature extraction, principal component analysis (PCA) for dimensionality reduction and a stacked ensemble (SVR-MLP-ridge regression) yielding the highest overall predictive accuracy (R² = 0.99, RMSE < 0.921 ppm, MAE < 0.808 ppm, EV = 0.99) across all analytes. The developed hybrid tree-PCA-ML framework enables interpretable frequency-based analysis consistent with the physicochemical interpretations from the equivalent electrical circuit models. This combined EIS-ML approach not only enhances predictive accuracy for nutrient concentrations but also identifies critical frequency regions governing the sensing mechanisms, offering a pathway toward high-precision, real-time water quality monitoring.