The heavy metals in the soil of an industrial agglomeration areas in Hebei Province were used to conduct a comparative study on the problem of characterization accuracy in handling non-stationary concentration and biased data with different interpolation methods including distance weighting（IDW）， ordinary Kriging（OK）， support vector machine（SVM）， and gradient enhanced decision tree（GBDT） to improve the accuracy of models for characterizing the soil pollution in industrial agglomeration areas. The results showed that the main pollutant in the soil of this site was arsenic， and GBDT exhibited the highest accuracy in characterizing arsenic pollution in soil （R²=0.911 5）. The results of GBDT visualization showed that the concentration distribution pattern of arsenic was "patchy aggregation" and had a good vertical migration capacity. The results of correlation analysis showed that the differentiation behavior of Arsenic concentration in the soil of this site was mainly related to the soil lithology and hydrogeological conditions. The results of Monte Carlo-based simulation showed that the total cancer risk index of both adults and children in the soil of this site exceeded the guidance value， and children suffered from higher non carcinogenic and carcinogenic risks than adults.