摘要： 在分类和预测任务中,包含大量不同取值的名词型属性使那些要求数值型输入的回归算法难以使用。 该文提出一种基于经验贝 叶斯统计方法的预处理方法,对这一类名词型属性进行变换,使之能用于预测建模。首先介绍了变换的统计学原理,然后给出了实现方法。分析表明,此方法简单易行,缩放性较好,而且在处理缺失数据时具有明显的优势。

关键词: 名词型属性, 预处理, 预测模型, 经验贝叶斯

Abstract: In classification and prediction task, nominal attributes characterized by a large number of distinct values represent a challenge for many classification and regression algorithms that require numerical inputs. This paper presents a preprocessing transformation method for nominal categorical attributes that allows this class of attributes to be used in predictive models, based on empirical Bayes statistics method. First, statistics principle of transformation is introduced, then the implementation is given. Analysis indicates this method is simple and scalable. Furthmore, the processing method is good at handling missing data.

中图分类号: 

• TP391.4