Abstract

Predicting cadmium (Cd(II)) adsorption in soils is critical for managing heavy metal contamination and mitigating its environmental risks. This study introduces a hybrid machine learning model that integrates Decision Trees (DT), Multi-Output Nonlinear Regression (MNLR), and Backpropagation Neural Networks (BPNN) to achieve accurate predictions of Cd(II) adsorption capacity. The model incorporates advanced data scaling techniques and feature expansion to effectively handle data heterogeneity and capture complex nonlinear relationships among soil properties, including Cation-Exchange Capacity (CEC), Organic Carbon Content (OC), clay content, pH, and soil-to-solution ratio. Sensitivity analysis identifies clay content as the most influential parameter, revealing its significant role in modulating adsorption behavior. The model demonstrates superior predictive performance, with an R² value of 0.898 and a substantial reduction in training loss, highlighting its potential for advancing environmental risk assessment and remediation strategies for contaminated soils. This work establishes a foundation for applying machine learning to optimize predictions in environmental science, offering insights into heavy metal adsorption and guiding the development of efficient remediation approaches.

Keywords:Machine learning; Heavy metal; Adsorption; Soil remediation; Decision tree; Neural network