rpartOrdinal: An R Package for Deriving a Classification Tree for Predicting an Ordinal Response

Area of Science:

• Statistics
• Machine Learning
• Computational Statistics

Background:

• Predicting ordinal responses in classification trees requires specialized methods.
• Standard impurity measures may not adequately capture the nature of ordinal data.
• Existing approaches may not account for varying misclassification costs.

Purpose of the Study:

• To introduce the rpartOrdinal R package for fitting classification trees with ordinal response variables.
• To implement and evaluate novel splitting functions tailored for ordinal data.
• To provide tools for analyzing ordinal associations within classification trees.

Main Methods:

• Implementation of generalized Gini impurity and Piccarreta's ordinal impurity functions.
• Inclusion of the ordered twoing splitting method.
• Development of a function for estimating the gamma statistic (ordinal measure of association).

Main Results:

• The rpartOrdinal package provides flexible and accurate methods for ordinal classification.
• The implemented splitting functions effectively handle the ordered nature of response categories.
• The package facilitates the incorporation of misclassification costs and association measures.

Conclusions:

• rpartOrdinal enhances the capability of classification trees for ordinal response prediction.
• The package offers valuable tools for statisticians and data scientists working with ordinal data.
• Accurate modeling of ordinal responses is crucial for reliable predictions in various fields.