Monitoring heavy metal(loid) concentrations in soils of industrially contaminated sites using machine learning models

Heavy metal(loid) contamination in soils at industrial sites represents a globally significant environmental concern. Traditional methods for monitoring soil heavy metal(loid) concentrations are limited by low efficiency, high costs, and time-consuming procedures. To address these limitations, machine learning (ML) has been increasingly adopted in recent years, given its powerful capabilities in nonlinear modeling and efficient data processing. This review systematically synthesizes 91 studies published between 2014 and 2025 that employed ML to monitor heavy metal(loid) concentrations in soils at contaminated sites. Analysis of the 91 selected studies reveals that Cu, Pb, As, Cd, Zn, Cr, Ni, and Hg are the most frequently investigated heavy metal(loid)s. Based on the types of input data used for ML, this review is structured into two sections using environmental covariates (35 studies) and hyperspectral data (56 studies). For environmental covariate-based monitoring, Random Forest (RF) is identified as the optimal model in 54.29 % of the reviewed studies. For hyperspectral data-based monitoring, Support Vector Machine (SVM), RF, and Partial Least Squares Regression (PLSR) are the most commonly employed modeling algorithms, each being used in at least 40 % of the reviewed studies. The application of ML-based monitoring of heavy metal(loid) concentrations in site soils includes predicting heavy metal(loid) concentrations, identifying drivers of heavy metal(loid) accumulation, characterizing spatial distributions, analyzing spatiotemporal evolution, and constructing three-dimensional (3D) distribution patterns. This review outlines the application workflow of ML, the optimal or frequently employed modeling algorithms, and representative application cases, while also discussing the current challenges and future perspectives in this field.