Enhancing water quality assessment in Skikda, Algeria using the PCA-based weighted index (WQI_P) and its predictive performance: a comparison with traditional WA_WQI approaches

Mostefa, Benacherine, Hinda, Hafid, Martínez, Alejandro, Noua, Allaoua, Abdelatif, Satour, Fadila, Fertas, Diganta, Mir Talas Mahammad, Nadjib, Lyazid Mohamed, Debassi, Bouchra, Uddin, Md Galal. (2026). Enhancing water quality assessment in Skikda, Algeria using the PCA-based weighted index (WQI_P) and its predictive performance: a comparison with traditional WA_WQI approaches. Journal of Contaminant Hydrology, 277, 104875. https://doi.org/10.1016/j.jconhyd.2026.104875

Abstract

Ensuring reliable river-water quality assessment is increasingly important in North Africa, where pollution pressures and data limitations complicate monitoring. Therefore, the research developed a principal-component-analysis–based water quality index (WQI_P) that is designed to address eclipsing, multicollinearity, and subjectively assigned weights that affect traditional indices such as the weighted-arithmetic WQI (WA_WQI). The objective of the research is to evaluate whether PCA-derived weights and objective parameter selection improve reliability, uncertainty, and classification stability. A dataset of 159 river-water samples from the Skikda region (Algeria) was analyzed. After screening correlated variables and extracting PCA contributions, WQI_P was constructed from the retained components. Eight machine-learning algorithms and a stacked ensemble were used under 10-fold cross-validation to compare the prediction performance and uncertainty of WQI_P and WA_WQI. Agreement metrics, PREI scores, confidence intervals, and class-transition analysis were used to assess the differences between the two indices, Predictive uncertainty was quantified using a Gaussian Monte Carlo simulation, which propagates variability by repeatedly perturbing model residuals to generate distributions of index predictions. The WQI_P consistently produced lower prediction errors (stacked RMSE = 2.74; MAE = 1.75) than the WA_WQI (RMSE = 3.16; MAE = 2.21), together with narrower 95% confidence intervals and reduced predictive uncertainty. The classification outcomes shifted toward a stricter and more balanced assessment: the proportion of samples classified as “Excellent” decreased (30 to 7), “Good” increased (55 to 88), and “Unsuitable” declined (40 to 12). These results indicated that grounding weights in the multivariate structure enhances stability and reduces dependence on a small set of dominant parameters. The findings demonstrated that the WQI_P can improve transparency, objectivity, and monitoring efficiency by focusing on the most informative variables. The index is applicable to data-scarce regions where objective weighting and uncertainty control are essential. Future work should test WQI_P across larger and more heterogeneous basins, extend validation using spatial–temporal blocking, and explore its integration into operational monitoring frameworks.

Publisher

Elsevier

Rights

CC BY

cb

Collections

Publications