원문정보
초록
영어
Decision tree induction has gained its popularity as an effective automated method for data classification mainly because of its simple, easy-to-understand, and noise-tolerant characteristics. The induced tree reveals the most informative attributes that can best characterize training data and accurately predict classes of unseen data. Despite its predictive power, the tree structure can be overly expanded or deeply grown when the training data do not show explicit patterns. Such bushy and deep trees are difficult to comprehend and interpret by humans. We thus propose a logic-based method to query over a complicate tree structure to extract only parts of the tree model that are really relevant to users’ interest. The implementation using ECLiPSe constraint language to perform constrained search over a decision tree model is given in this paper. The illustrative examples on medical domains support our hypothesis regarding simplicity of constrained tree-based patterns.
목차
1. Introduction
2. Building a Decision Tree Model with Logic Programming
3. Querying a Tree Model
4. Experimentation and Results
5. Conclusion
Acknowledgements
Appendix
References