Improving Classification Trees with Simulated Annealing,
Abstract
Classification trees produced by a recursive partitioning scheme such as CART are not guaranteed to be the best tree structured classifiers possible, partly because the sequential manner by which they are formed does not allow for looking ahead . In some cases, altering trees produced by CART by shifting the partition boundaries results in improved prediction rules. Simulated annealing can be used to efficiently search for trees which may perform better than those produced by CART.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 1992
- Accession Number
- ADP007175
Entities
People
- Clifton D. Sutton
Organizations
- George Mason University