statistics - What are the ways of pre-processing categorical data before applying classification algorithms?

Question

I am new to machine learning and I am working on a classification problem with Categorical (nominal) data. I have tried applying BayesNet and a couple of Trees and Rules classification algorithms to the raw data. I am able to achieve an AUC of 0.85.

I further want to improve the AUC by pre-processing or transforming the data. However since the data is categorical I don't think that log transform, addition, multiplication etc. of different columns will work here.

Can somebody list down what are most common transformations applied on categorical data-sets? ( I tried one-hot encoding but it takes a lot of memory!!)

Answer 1

正如您所提到的，根据我的经验，分类是最好的处理单热编码（例如转换为二进制向量）。如果内存是一个问题，那么使用在线分类算法并动态生成修改后的向量可能是值得的。

除此之外，如果类别代表一个范围（例如，如果类别代表一系列值，如年龄、身高或收入），则可以处理中心（或一些适当的平均值，如果有标签内分布）的类别范围作为一个实数。

如果您正在应用聚类，您还可以将分类标签视为轴上的点（1、2、3、4、5 等），并适当地缩放到其他特征。