Abstract: In order to solve the problem of site classification jump change of thick overburden layer caused by the influence of test environment, instrument type, personnel experience and other factors on reading equivalent shear wave velocity, this paper collects a large number of relevant field test data of thick overburden layer, uses machine learning method to conduct training modeling, and then studies the site classification jump change problem under thick overburden layer. The results show that the classification accuracy of the random forest model reaches 97.7% after adding the "equivalent coefficient of variation", and its generalization ability and the overall judgment ability of the sample are better than those of the support vector machine model. This model provides a new way to judge the type of construction sites with thick overburden. By comparing the results of the second judgment with the results of the exploration report, it is proved that the random forest model can be used for the second judgment of the jumping change problem of site classification, which provides a reliable basis for avoiding overly conservative judgment in similar situations in the project site.