Integration of advanced chemometric and machine learning techniques: Stacking model and XGBoost dynamic correction for aniline detection with an unmodified carbon paste electrode

This study proposes a novel Chemometric approach for the precise and reliable quantification of aniline using an unmodified carbon paste electrode (CPE), even in the presence of interfering species such as phenol. The proposed methodology integrates partial least squares (PLS), random forest (RF) and ridge regression in a stacked model framework, combined with particle swarm optimization (PSO) for potential selection. To enhance the prediction accuracy, a dynamic correction based on XGBoost was applied, which effectively minimized residual errors and improved model robustness. The model developed demonstrated high performance on the training (R² = 0.9993, RMSE = 0.6260 and MAE = 0.4773), test (R² = 0.9979, RMSE = 0.7616 and MAE = 0.6839) and cross-validation datasets (R² = 0.9988, RMSE = 0.8312 and MAE = 0.5467), thus confirming its stability and reliability. The high coefficient of determination (R2), as well as the low root mean square error (RMSE) and mean absolute deviation (MAE) values, underscore the model's robust predictive capabilities and its capacity to accommodate intricate electrochemical interactions. Moreover, the calculated recovery rates and tolerance limits signify high precision and consistency across varying concentration levels. Utilizing an unmodified EPC, this approach proffers a cost-effective, sensitive, and selective strategy for aniline quantification, rendering it especially well-suited for environmental and industrial monitoring. The model's strong predictive capabilities underscore its potential for real-world contaminant detection, offering a reliable solution for complex analytical environments.