基于决策树的区域地块风险管控模式预测

针对传统地块风险管控模式确定存在效率低、成本高、缺乏系统性的缺陷，通过构建区域地块风险管控模式决策的特征数据集，并采用卡方自动交叉检验（CHAID）、穷举卡方自动交叉检验（E-CHAID）、分类与回归树（CART）3种决策树（DT），探索不同DT算法应用于区域地块风险管控模式预测的可行性.结果表明，DT应用于区域地块风险管控模式预测是可行的.CART-DT在准确率（ACC）、精度（PRE）、召回率（REC）、F₁值方面的性能均显著优于CHAID-DT和E-CHAID-DT.CART的总体优化算法可能更适合区域地块风险管控模式的预测.区域保护目标（RPG）、区域污染物类型（RPT）、区域企业平均生产年限（RAPP）3项输入变量对CART-DT输出的重要性非常高；区域年平均风速（RAAWS）、区域地形地貌（RT）、区域土地增值潜力（RLVP）等11项输入变量对CART-DT输出的重要性较高；区域人口密度（RPD）、区域主导行业风险（RDIS）等6项输入变量对CART-DT的输出也有一定贡献.

Prediction performance of risk management and control mode in regional sites based on decision tree

In order to overcome the defects of low efficiency, high cost and lack of systematicness, which had been found in screening of traditional risk management and control mode in regional sites. Three decision tree (DT) algorithms, including Chi-squared Automatic Interaction Detector (CHAID), Exhaustive CHAID (E-CHAID), and Classification And Regression Tree (CART), were implemented to predict risk management and control mode in regional sites. Characteristic data set was constructed using existing risk management and control mode of regional sites and relevant regional attributes. The results showed that, DT-based mode could be used to predict risk management and control mode in regional sites. In the aspects of accuracy (ACC), precision (PRE), recall ratio (REC) and F₁ value, the performance of CART-DT was superior to CHAID-DT and E-CHAID-DT. The overall optimization algorithm of CART might be more suitable for predicting risk management and control mode in regional sites. Regional protection goal (RPG), regional pollution type (RPT) and regional average production period (RAPP) were the three dominant factors of CART-DT output. Eleven input variables such as regional average annual wind speed (RAAWS), regional topography (RT) and regional land value-added potential (RLVP) were relatively important to CART-DT output. Regional population density (RPD), regional dominant industry risk (RDIS) and four other input variables had relatively low impact on CART-DT output.