Prediction of cadmium content using machine learning methods

Abstract Heavy metals are the most environmentally hazardous pollutions in agricultural soils, threatening humans and several ecosystem services. Cadmium (Cd) is a highly toxic element but distinctively different from other heavy metals with its high mobility in soil environments. The study aimed to evaluate the Cd concentration of soils in the Konya plain with a specific attribute to soil fertilization, mainly phosphorous fertilizers. A total of 538 surface (0–20 cm) soil samples were analyzed to determine basic physical and chemical properties and total phosphorus (P) and Cd concentrations. Descriptive statistics, machine learning, and regression models were used to assess the accumulation of Cd in soils. Decision Trees, Linear Regression, Random Forest, and XGBoost machine learning methods were used in Cd prediction. The XGBoost model proved to be the best prediction model, with a coefficient of determination of 98.1%. Electrical conductivity, pH, CaCO 3 , silt, and P were used in the Cd estimation of the XGBoost model and explained 56.51% of the total variance in relation to measured soil properties. The results revealed that a machine learning algorithm could be useful for estimating Cd concentration in soils using basic physical and chemical soil properties.