X‐ray fluorescence (XRF) analysis is exceedingly suitable for detecting heavy metal contents in soil. In order to do that, an accurate prediction model based on XRF analysis is necessary. But in practice, the XRF spectral data is susceptible to moisture content in soil, which may lead to inaccurate prediction results. Accordingly, a new prediction model based on Random Forest Regression (RFR) and improved Sparrow Search Algorithm (SSA) was proposed, which takes the variation of moisture content into consideration. At first, the XRF spectral data were obtained by experiment. Owing to the advantages of training speed and prediction ability, the RFR was employed to predict the heavy metal contents. In order to further improve the performance of RFR, the SSA was selected and improved with theory of good‐point set, which can determine optimum hyper‐parameters of RFR conveniently. It can be found by comparison that the proposed model outperforms other commonly used models.

In order to extract useful information from XRF (X-ray fluorescence) spectrum, and establish a high-accuracy prediction model of soil heavy metal contents, a hybrid model combined Deep Belief Network (DBN) with tree-based model was proposed. The DBN was firstly introduced into feature extraction of XRF spectral data, which can obtain deep layer features of spectrum. Owing to the strong regression ability of tree-based model, it can offset the deficiency of DBN in prediction ability, so it was used for predicting heavy metal contents based on the extracted features. In order to further improve the performance of the model, the parameters of model can be optimized according to the prediction error, which was completed by sparrow search algorithm (SSA) and the gird search. The hybrid model was applied to predict the contents of As and Pb based on spectral data of overlapping peaks. It can be obtained that R2 of As and Pb reached 0.9884 and 0.9358, the MSE of As and Pb are as low as 0.0011 and 0.0058, which outperform other commonly used models. That proved the combination of DBN and tree-based model can obtain more accurate prediction results.