结合随机森林与地理探测器的村域贫困分布格局及其分异机制分析

Analysis on the Distribution Pattern and Differentiation Mechanism of Rural Poverty by Combining Random Forest and Geographical Detector

乡村贫困是我国经济发展过程中的重要问题,很多相关研究从定性或定量角度出发,分析地方或区域贫困问题,对扶贫脱贫的理论及实践产生了重要的意义。运用定量研究方法,结合机器学习与地理探测器分析村级贫困问题。以广东省重点扶贫地区连州市星子镇为例,选取12个对星子镇的贫困发生率产生影响的潜在因素,基于随机森林算法和地理探测器,对该镇21个行政村贫困分异的影响因素及其影响程度进行分析。结果表明:随机森林算法与地理探测器由于原理的差异,在分析各因子影响力这同一目的的前提下,得出数据结果存在差异;星子镇的贫困分异是多因子间的共同正向促进作用,具有复杂性,致贫需多模式综合并行;对于星子镇这个自然约束较大的地区,脱贫更需要着重降低社会阻隔,通过政策、资源的倾斜减缓社会阻隔,从而进一步促进自然约束力的降低,最终达到脱贫目的。

Rural poverty is an important issue in the process of China’s economic development.Many related studies analyze the local or regional poverty issues from a qualitative or quantitative perspective,which has important implications for the theory and practice of poverty alleviation.This article mainly used quantitative research methods,combined with machine learning and geographic detectors to analyze village-level poverty.Taking Xingzi Town,Lianzhou,a key poverty alleviation area in Guangdong Province as an example,and selecting 12 potential factors that affect the incidence of poverty in Xingzi Town,based on random forest algorithm and geographic detectors,the influencing factors and degree of poverty differentiation of 21 administrative villages in Xingzi Town,Lianzhou City were analyzed.The results indicated:due to the difference in principle between the random forest algorithm and the geographic detector,on the premise of analyzing the influence of various factors,the results are different;poverty differentiation in Xingzi Town is a common positive promoting effect among multiple factors,which means poverty differentiation is complex,and multi-mode integration is needed to get rid of poverty;for Xingzi Town,a region with greater natural constraints,poverty alleviation needs to focus on reducing social barriers.Through the tilting of policies and resources,social barriers are slowed down,which further promotes the reduction of natural constraints and ultimately achieves the goal of poverty reduction.