林火数据的Logistic和零膨胀Poisson(ZIP)回归模型

Logistic and ZIP Regression Model for Forest Fire Data

利用统计模型对森林火灾数据进行了描述和分析,所用模型包括Logistic回归模型和零膨胀Poisson(ZIP)回归模型。将所用的森林火灾数据分别视为分组因子数据和有序变量数据进行建模。为了进行预测和验证,建模时使用部分数据,其余数据作为检验数据,用以检验预测的准确性。研究结果显示,所研究的两类模型都得到了与检验样本接近的结果,具有较好的火灾次数预测能力。其中零膨胀模型不仅可以得到与Logistic模型相当的结果,而且能够有效解决火灾数大于天数的问题,以及可以对零值过多的数据进行较好地建模。该研究表明所建立的Lo-gistic回归模型和零膨胀Poisson回归模型都是分析火灾与影响因素相关性的适用模型。

In this paper, we use Logistic and Zero-Inflated Poisson （ZIP） regression models to describe and analyze the forest fire data. If we regard the forest fire data as the grouped factor data and ordinal state variable data respectively, then we can model the forest fire data by the Logistic and ZIP regression models. We use one part of forest fire data for modelling, and the rest are used to predict and validate the veracity of models. Then, the best Logistic models and the best ZIP regression models are selected. The prediction results show that the selected Logistic and ZIP regression models can be applied to analyze the relationship between fire and its influence factors.