Integrated multivariate feature selection and hydrochemical analysis for magnesium assessment in a conventional water...

Magnesium (Mg²⁺) is an important water-quality parameter influencing human health, water hardness, and hydrochemical stability in treatment systems. This study investigated the factors controlling Mg²⁺ concentration at the Tamburawa Water Treatment Plant (TWTP), Kano, Nigeria, using an integrated statistical and multivariate framework. The dataset comprised ten physicochemical parameters: aluminum (Al), chloride (Cl^-), fluoride (F^-), iron (Fe), manganese (Mn), nitrite (NO₂^-), nitrate (NO₃^-), sulfate (SO₄^2-), total suspended solids (TSS), and Mg²⁺. Stationarity was assessed using the Augmented Dickey-Fuller (ADF) and Phillips-Perron (PP) tests, which confirmed that all variables, including Mg²⁺, were stationary at level form, I(0). Pearson correlation analysis showed weak linear associations between Mg²⁺ and most explanatory variables, indicating that Mg²⁺ variability is not governed by simple pairwise relationships. To further evaluate variable influence, one-way analysis of variance (ANOVA), principal component analysis (PCA), minimum redundancy-maximum relevance (mRMR), and sensitivity analysis were applied. ANOVA confirmed statistically significant differences among the studied parameters (F = 191.07, p < 0.001), while PCA showed that the first four components explained approximately 61.6% of the total variance. Feature-ranking results identified Cl^-, NO₃^-, and NO₂^- as the most informative variables associated with Mg²⁺ variability. Sensitivity analysis further showed that Cl^- and NO₃^- had the strongest influence, with sensitivity scores of 1.5698 and 0.9493, respectively. The findings indicate that Mg²⁺ behavior at TWTP is controlled primarily by dissolved ionic-strength and nutrient-related hydrochemical gradients rather than by suspended solids or trace-metal variability. Overall, the study provides a statistically validated framework for Mg²⁺-centered monitoring by identifying stable, influential, and non-redundant hydrochemical predictors in a conventional surface water treatment system.