Magnetic Properties as a Proxy for Predicting Fine-Particle-Bound Heavy Metals in a Support Vector Machine Approach

Abstract

The development of a reasonable statistical method of predicting the concentrations of fine-particle-bound heavy metals remains challenging. In this study, daily PM samples were collected within four different seasons from a Chinese mega-city. The annual average PM concentrations determined in industrial, city center, and suburban areas were 90, 81, and 85 μg m, respectively. Environmental magnetic measurements, including magnetic susceptibility, anhysteretic remanent magnetization, isothermal remanent magnetization, hysteresis loops, and thermomagnetism, indicated that the main magnetic mineral of PM is low-coercivity pseudosingle domain (PSD) magnetite. Using a support vector machine (SVM), both the volume- and mass-related concentrations of heavy metals were predicted by the PM mass concentrations and meteorological factors, with or without magnetic properties as input variables. The inclusion of magnetic variables significantly improved the prediction results for most heavy metals. Predictions based on models that included the magnetic properties of the metals Al, Fe, Mn, Ni, and Ti were promising, with R values of >0.8 in both the training and the test stages as well as relatively low errors. Our results demonstrate that the inclusion of environmental magnetism in a SVM approach aids in the effective monitoring and assessment of airborne heavy-metal contamination in cities.

Influent pollutant monitoring in decentralised treatment facilities using electrical conductivity sensors: experimental and modelling insights

“…The algorithm has been proven effective in handing various nonlinear classification problems using the kernel function to map input variables into a high-dimensional feature space, after which the nonlinear feature mapping allows for the treatment of nonlinear problems in a linear space. 31,32 The input of the model was a [308 × 7] matrix that was composed of 308 rows of water quantity data (TN, TP, Na + , K + , Ca 2+ and Cl − concentrations), whereas the output was a [308 × 1] matrix that was composed of the corresponding EC values. A total of 278 rows were randomly selected to be the training set, whereas the remaining approximately 10% of the data (30 rows) were placed in a testing set.…”

Biomagnetic monitoring combined with support vector machine: a new opportunity for predicting particle-bound-heavy metals

“…Among those methods, the support vector machine (SVM) algorithm, which is based on the structural risk minimization principle, has increasingly been applied to solve non-linear regression problems, because it takes into account the error approximation in the data and generalization of the models. Previous studies used SVM approaches to predict a series of atmospheric pollutants, including NO 2 40 , CO 41 43 , based on emission information and meteorological data. However, the potential of statistical models combined with leaf magnetic properties to predict atmospheric heavy metals has yet to be fully explored.…”

A new methodological framework for geophysical sensor combinations associated with machine learning algorithms to understand soil attributes

“…Recently, many models have been used to estimate soil attributes and their spatial distribution from geophysical data (gamma ray, κ, and ECa) and soil attributes, including machine learning algorithms, such as the support vector machine (SVM; Priori et al, 2014;Heggemann et al, 2017;Li et al, 2017;Leng et al, 2018;Zare et al, 2020), random forest (Lacoste et al, 2011;Viscarra Rossel et al, 2014;Harris and Grunsky, 2015;Sousa et al, 2020), KNN and artificial neural network (ANN) (Dragovic and Onjia, 2007), and Cubist (Wilford and Thomas, 2012) methods.…”