基于交互效应Logistic回归模型的耕地质量评价方法研究

Evaluation of Cultivated Land Quality Based on Interactive Logistic Regression Model

在以往的耕地质量评价工作和研究中,存在对监测评价指标考虑不周、评价方法滞后、主观性强等问题。文章旨在探寻一种准确、高效的耕地质量评价方法,挖掘耕地质量影响因素间的交互效应,构建基于数据驱动的耕地质量评价指标体系,避免确定指标权重,同时提高评价效率。以从化区耕地为研究对象,使用分层抽样法抽取不同等别的耕地质量样本,共计6 000组,利用关联规则方法寻找数据间的关联关系,得到耕地质量评价指标间可能的交互效应,并结合Logistic回归模型对耕地质量进行评价。结果表明,当关联规则的最小支持度阈值分别设置为0.05、0.03和0.01时,挖掘出关联规则数分别为8、35、78条,涉及到的交互效应个数分别为6、32、75个,结合模型参数分析,其中具有统计学意义的交互效应有5个,分别是:耕地利用方式、表层土壤质地、有效土层厚度对土壤酸碱度的交互效应,地形坡度对土壤有机质的交互效应,灌溉保证率对排水条件的交互效应。将评价指标和指标间的交互效应均纳入Logistic回归模型,该模型能获得92.2%的耕地评价正确率,达到较高的精度等级。关联规则分析与Logistic回归模型相结合,获得影响耕地质量评价指标间的交互效应,是一种行之有效的方法,为耕地质量驱动因素间交互效应的发现提供了有益线索,该模型对比全局Logistic回归模型具有更高的模型拟合优度和预测准确率,可满足实际应用需求。

There are some problems such as incomplete evaluation indexes,the strong subjectivity and lagging methods in cultivated land quality evaluation work and research.The purpose of this paper is to explore an accurate and efficient evaluation method of cultivated land quality evaluation,excavate the interaction among the influencing factors of cultivated land quality,construct a datadriven evaluation index system for cultivated land quality which can avoid determining index weights and improve evaluation efficiency.This method can avoid setting index weight and improve the evaluation efficiency.Taking cultivated land in Conghua,Guangzhou as the research object,6000 groups of cultivated land quality samples of different grades were extracted by stratified sampling method.The association rules method was used to find the possible interaction effect among cultivated land quality grading indexes.Finally,evaluating the quality of cultivated land by interactive Logistic regression model.The result shows that association mining can effectively find the interaction effect among cultivated land quality grading indexes.When the minimum support degree is set to 0.05,0.03 and 0.01,the number of association rules is 8,35 and 78,and the number of interaction effects involved is 6,31 and 75 respectively.Combined with the model parameter analysis,there are 5 statistically significant interactions,which are:the interaction of cultivated land use type,surface soil texture and effective soil thickness on soil pH,the interaction of terrain slope on soil organic matter,and the interaction of irrigation guarantee rate on drainage system.The cultivated land quality evaluation result shows that 92.2%of cultivated land evaluation accuracy can be obtained by integrating the grading index and the interaction effect among the indexes into the Logistic regression model.The combination of association rule analysis and Logistic regression model is an effective method to obtain the interaction among cultivated land quality evaluation indexes,which provides useful clues for the discovery of interaction among driving factors of cultivated land quality.Compared with the global Logistic regression model,the model has higher accuracy rate of cultivated land quality evaluation and can meet the needs of practical application.