基于机器学习方法的农业转移人口市民化水平影响因素研究

Factors Influencing the Citizenization of Rural-Urban Migrants:A Study Based on Machine Learning

本文基于2017年中国流动人口动态监测调查数据,系统地运用多元线性回归、惩罚回归、集成学习和深度学习等多种机器学习方法,考察了农业转移人口市民化水平的影响因素。研究结果表明:集成学习方法在预测农业转移人口市民化水平方面明显优于多元线性回归模型,其中梯度提升回归树模型的预测效果最佳;在所有特征变量中,个体的受教育程度、性别、家庭规模、年龄和流动城市数量是影响农业转移人口市民化水平的最主要因素。此外,本文通过累积局部效应图展示了不同影响因素对农业转移人口市民化水平的具体预测模式,并发现年龄和流动城市数量对农业转移人口市民化水平有着明显的非线性影响。这些研究结论对政府进一步推进农业转移人口市民化具有重要参考价值。

Based on the China Migrants Dynamic Survey(CMDS)data in 2017,this paper systematically investigates the influencing factors of the citizenization of rural-urban migrants using the multiple linear regression,penalized regression,ensemble learning,deep learning,and other machine learning methods.The results show that the ensemble learning method outperforms the multiple linear regression method in predicting the level of citizenization of rural-urban migrants,among which the gradient boosting regression tree(GBRT)model performs the best prediction.Moreover,among all the characteristic variables,the individual's education level,gender,family size,age,and number of mobile cities are the most important influencing factors for the citizenization of rural-urban migrants.In addition,we use the accumulated local effects(ALE)plot to show the specific prediction patterns of different influencing factors,and find that factors such as age and number of mobile cities have non-linear characteristics in their influence on the citizenization of rural-urban migrants.The conclusions of this paper are informative for the design of policies aimed at accelerating the citizenization of rural-urban migrants.