Multi-factor interaction perspective: machine learning-based analysis of Ni2⁺ adsorption onto soil.

Abstract

With the rapid development of industry and agriculture, the ecological and health impacts of nickel (Ni²⁺) have gained increasing attention. While previous experimental studies have identified factors influencing Ni²⁺ adsorption behavior in soils, their nonlinear relationships and interactive effects remain underexplored. Through combining machine learning (CatBoost/XGBoost) models with SHapley Additive exPlanations (SHAP), this study analyzed 662 experimental datasets to reveal these nonlinear interactions between factors that affect the adsorption behavior of Ni²⁺ in soil. The modeling results demonstrated CatBoost's superior performance over XGBoost (test R² = 0.85 vs 0.83). Both feature importance analysis from the model and SHAP values identified the initial Ni²⁺ concentration (C₀) as the most critical factor, followed by ionic strength (IS), solid-to-liquid ratio (SL), clay content, and cation exchange capacity (CEC). SHAP dependence plots revealed a nonlinear SL effect that maximum adsorption occurred at low SL ratios with subsequent fluctuations attributable to ionic competition and pore accessibility constraints. Notably, SHAP interaction analysis uncovered a key finding which C₀ exhibited synergistic interactions with both CEC and clay content to enhance Ni²⁺ immobilization, whereas elevated IS substantially diminished these cooperative effects. This work quantitatively characterizes multifactorial coupling in Ni²⁺ adsorption processes, advancing theoretical foundations for risk assessment while informing targeted remediation strategies and enhancing mechanistic understanding of heavy metal interactions in soil systems.