基于随机森林模型的分类数据缺失值插补

Missing Data Imputation for Categorical Data Based on Random Forest Model

缺失数据是影响调查问卷数据质量的重要因素,对调查问卷中的缺失值进行插补可以显著提高调查数据的质量。调查问卷的数据类型多以分类型数据为主,数据挖掘技术中的分类算法是处理属性分类问题的常用方法,随机森林模型是众多分类算法中精度较高的方法之一。将随机森林模型引入调查问卷缺失数据的插补研究中,提出了基于随机森林模型的分类数据缺失值插补方法,并根据不同的缺失模式探讨了相应的插补步骤。通过与其它方法的实证模拟比较,表明随机森林插补法得到的插补值准确度更优、可信度更高。

Missing data is a important factor which has bad effect on the data quality of survey questionnaire, missing data imputation can significantly improve the data quality. Categorical data is the main data type of survey data. Classification algorithms of data mining can be often dealt with classification problem, random forest modeling is one of the high predictive accuracy classification models. This paper introduces the random forest model into the missing data imputation research of survey data, and proposes the missing data imputation method for categorical data based on random forest model. Imputation process is also designed according to different pattern of missing data. Empirical simulation shows that the proposed new method can obtain more accuracy and reliable results by comparing with other imputation methods.