基于Logistic模型的山区县域村分类方法研究 — —以重庆市巫溪县为例

为了实现乡村振兴战略对村分类（搬迁撤并、城郊融合、集聚提升和特色保护）的要求，以典型的山地县域重庆市巫溪县为例，从生态保护、社会经济和历史人文等指标体系中筛选12个指标作为自变量，构建多分类Logistic回归模型，快速精准识别村分类。以通过评审的村布局方案中随机选取170个村作为实验样本，以特色保护类为参照组，回归模型的拟合优度高达0.911。利用模型对剩余的119个村类型进行预测，搬迁撤并、城郊融合、集聚提升、特色保护四种分类的正确率分别为90.57%、86.36%、88.89%、82.35%，对预测概率较低（<75%）的个别样本加以人工判断，可得到较理想的分类结果。该方法可以对村进行快速初分类，为人工判别与调整提供了初始分类方案，提高了村分类的科学性和工作效率，弥补了当前村分类方法的不足。

A Classification Method of Mountainous Villages Based on Logistic Model: A Case Studyon Wuxi County, Chongqing Municiparity

In the context of rural revitalization and territorial spatial planning, it is animportant step to divide a large number of villages into different types before makingvillage planning. In order to identify types of villages quickly and accurately inmountainous area, taking administrative village (total 289) in Wuxi county in ChongqingMuniciparity as an example, a multiple logistic regression model is established in whichthe independent variables are four types (relocation and merging, urban-suburbanintegration, agglomeration for improvement, protection due to special characteristics) andthe dependent variables are 12 indicators including ecological protection, social economy,history and culture, base on 170 villages randomly selected from the approved villagelayout scheme. In the model, the reference group is protection due to special characteristicstype. SPSS25.0 software is used to construct the multi classification logistic regressionmodel and calculate the fitting coefficient of the regression model.The results show that Nagelkerke R 2 =0.911, Cox & Snell R 2 =0.839, McFadden R 2 =0.721,the comprehensive accuracy of regression model is as high as 91.20%, and the fitting effectis good. The results of the relocation and merging model show that the proportion of thearea covered by the ecological protection red line, the total economic volume, the villagelevel, the scale of the village construction land, the distance from the center of the county,the scale of historical cultural sites, and whether the village is a characteristic tourist villagehave a significant impact. The results of the urban-suburban integration model show that thetourism income, the village level, the scale of the village construction land, the distance fromthe center of the county, the road network density, the scale of historical cultural sites, andwhether the village is a characteristic tourist village have a significant impact. The resultsof the agglomeration for improvement model show that the proportion of the area covered by the ecological protection red line, the tourism income the village level, the population potential, the scale of the village construction land, thescale of historical cultural sites, the scale of rural tourism projects, and whether the village is a characteristic tourist village have a significantimpact. Then the model is used to predict the remaining 119 administrative villages, the accuracies of the four types of relocation and mergingor urban-suburban integration, agglomeration for improvement and protection due to special characteristics are 90.57%, 86.36%, 88.89% and82.35% respectively, and the comprehensive correct rate is as high as 87.97%. Then the model is used to predict 289 administrative villages, andthe comprehensive correct rate is as high as 89.97%.The main reason for the misclassification is the homogenization of village characteristicindexes. Compared with the correct classification of villages, the indicators of the misclassified villages are not prominent, which can beaccurately classified only by combining with the actual situation and assisted by artificial judgment. However, the prediction and classification ofthe model are completely based on the rules of data, which leads to the inconsistency between the predicted village classification and the actualclassification. According to the simulation prediction probability, only 12 villages with a probability of less than 75% (total 52) are misclassified,and only 2 villages with a probability of less than 50% (total 8) are misclassified. It can be seen that most of the classifications predicted by themodel are in line with reality. Combined with the characteristics of the county and the spatial distribution rules of each classification, the idealclassification result can be obtained by manually judging several samples with low prediction probability (less than 75%) based on the predictionof the model.The accuracy of the multiple logistic regression model depends on the comprehensiveness of the indicators and the number of randomlyselected typical samples. The perfect index system and the representative index with low correlation are more conducive to the construction ofstable model. According to the actual situation, the selection of regional characteristics index can make the classification results more in line withthe reality. In addition, the village classification process is complex, and there are few models which can be taken into account all the influencingfactors, including some factors that are difficult to quantify. Therefore, further manual subdivision is necessary. In general, this method combinessufficient quantitative analysis and a small amount of artificial judgment, which can greatly reduce the workload and subjectivity of traditionalclassification of villages. It can not only be applied in village classification in mountainous area, but also can be applied in plain and hilly areawith adjusting indicators according to local characteristics, which is conducive to improving the scientificity and work efficiency of villageclassification. It is necessary for all regions to modify and coordinate according to the actual situation.