摘要：In this paper, a classification model for the mi- gration of residents without local household registration in Beijing is established through the algorithm of Support Vector Machine (SVM) and the model is verified by using the migration data of Beijing, which is collected from various surveys. Our result shows that, compared to BP Neural Network and Logistic Regression, SVM performs better in terms of accuracy and generalization for these particular classification tasks. We identify ten classification features, which, we believe, are crucial as the determining factors to predict the migration trend in Beijing. These ten features include age, education, occupation, income, family status, housing status, leisure status, insurance status, temporary residence permit status and residence time. Our research shows that, taking into account the population demographic attributes and behavioral characteristics, our SVM classification model is able to predict the migration trend with a high accuracy rate. We believe that the results presented in this paper will provide valuable practical insights for various governmental departments of megacities in grasping the migration trend of different types of residents without local household registration, as well as in improving the residence policies, in order to encourage outward migration and tackle the issue of rapid population growth.
关键词：megacity; residents without local household registration; migration; support vector machine.
