Classification and Regression Trees for Symbolic Data

Compared to classical data which take a single value, there is another type of data, symbolic data, which can be a list, an interval, and even a distribution into consideration. Symbolic data are very common in our daily life; however, the analysis methods for symbolic data are very limited. For instance, a famous and useful method for supervised learning such as regression or classification is the decision tree. There are many useful algorithms based on the decision tree. However, the decision tree is only useful to classic data taking a single value, either numerical or categorical. In this dissertation, I will extend the classification and regression tree method (CART) to symbolic data.