A Hybrid Regression Model for Mixed Numerical and Categorical Data

Abstract

It is noticeable in different heterogeneity types that complexity is inherent in heterogeneous data, and regression analysis methods are well defined and exhibit high-accuracy performance with numeric data. However, real-world problems contain non-numerical variables. There are two main approaches to handling mixed-type data sets in regression analyses. The first approach is unifying data types for all the variables (such as continuous numerical data) and then applying the regression analysis. However, this approach degrades the data quality, as some original data types are converted to other types in the learning stage. The second approach is to apply some similarity measurements, which can be highly complex in some situations. To overcome these limitations, we propose a tree-based regression model to effectively handle the mixed-type data sets without using a dummy code or a similarity measurement.