基于随机森林模型的东北地区收缩城市分布格局及影响因素研究

基于2005—2009年、2010—2014年、2015—2019年和2005—2019年人口变化数据，判定东北地区收缩城市，分析其空间分布格局，并对比运用多元线性回归方法和随机森林回归方法探索东北地区收缩城市形成的影响因素及其影响作用。结果表明：① 空间上，东北地区的收缩城市主要分布在以长白山、三江平原、小兴安岭和大兴安岭等地区为代表的“陆上边缘”地区；时间上，收缩中心呈现出明显向北移动的态势，与之相对的扩张中心呈现出向南移动的态势，并且收缩城市进一步集聚；② 在影响因素的研究上，多元线性回归与随机森林回归结果都表明社会经济因素对收缩城市形成起了主要作用；③ 随机森林回归的精度比多元线性回归高，其结果显示人均GDP对收缩强度的影响程度最大，随后依次为失业率、科学教育事业费、在岗职工平均工资，且这4个影响因素中仅失业率对收缩起促进作用，其余3个影响因素在不同程度上抑制收缩城市的形成。

Shrinking Cities Distribution Pattern and Influencing Factors in Northeast China Based on Random Forest Model

Based on the demographic data of four periods (2005-2009, 2010-2014, 2015-2019 and 2005-2019), shrinking cities were identified in Northeast China, and their spatial distribution patterns were analyzed. We further compared the multiple linear regression (MLR) and the random forest regression (RFR) to explore the influencing factors and mechanisms of shrinking cities in Northeast China. The results showed that: 1) Spatially, the shrinking cities in Northeast China were mainly distributed in the "land-edge" regions represented by the Changbai Mountains, the Sanjiang Plain, Lesser and the Da Hinggan Mountains. Temporally, the shrinkage center showed an obvious northward trend, while the expansion center showed a southward trend. In addition, the shrinking cities were further clustered. 2) The results of both MLR and RFR indicated that socio-economic factors play a major role in the formation of shrinking cities. 3) The accuracy of RFR was higher than that of MLR. The results of RFR showed that GDP per capita has the greatest influence on the shrinkage intensity, followed by unemployment rate, expenses of science and education, and average wage of employed workers. There four influencing factors, except unemployment rate, the remaining three influencing factors restrict the formation of shrinking cities to varying degrees.