Comparison between machine linear regression (MLR) and support vector machine (SVM) as model generators for heavy metal assessment captured in biomonitors and road dust

Exposure to suspended particulate matter (PM), found in the air, is one of the most acute environmental problems that affect the health of modern society. Among the different airborne pollutants, heavy metals (HMs) are particularly relevant because they are bioaccumulated, impairing the functions of living beings. This study aimed to establish a method to predict heavy metal concentrations in leaves and road dust, through their magnetic properties measurements. For this purpose, machine learning, automatic linear regression (MLR), and support vector machine (SVM) were used to establish models for the prediction of airborne heavy metals based on leaves and road dust magnetic properties. Road dust samples and leaves of two common evergreen species (Cupressus lusitanica/Casuarina equisetifolia) were sampled simultaneously during two different years in the Great Metropolitan Area (GMA) of Costa Rica. MLR and SVM algorithms were used to establish the relationship between airborne heavy metal concentrations based on single (χlf) and multiple (χlf y χdf) leaf magnetic properties and road dust. Results showed that Fe, Cu, Cr, V, and Zn concentrations were well-simulated by SVM prediction models, with adjusted R² values ≥ 0.7 in both training and test stages. By contrast, the concentrations of Pb and Ni were not well-simulated, with adjusted R² values < 0.7 in both training and test stages. Heavy metal predicción models using magnetic properties of leaves from Casuarina equisetifolia, as collectors, yielded better prediction results than those based on the leaves of Cupressus lusitanica and road dust, showing relatively higher adjusted R² values and lower errors (MAE and RMSE) in both training and test stages. SVM proved to be the best prediction model with variations between single (χlf) and multiple (χlf y χdf) magnetic properties depending on the element studied.