Semester

Fall

Date of Graduation

2005

Document Type

Thesis

Degree Type

MS

College

Statler College of Engineering and Mineral Resources

Department

Lane Department of Computer Science and Electrical Engineering

Committee Chair

Bojan Cukic.

Abstract

Machine learning knowledge representations, such as decision trees; are often incomprehensible to humans. They can also contain errors specific to the representation type and the data used to generate them. By combining larger; less comprehensible decision trees, it is possible to increase their accuracy as an ensemble compared to the best individual tree. The thesis examines an ensemble learning technique and presents a unique knowledge elicitation technique which produces an ordered ranking of attributes by their importance in leading to more desirable classifications. The technique compares full branches of decision trees, finding the set difference of shared attributes. The combination of this information from all ensemble members is used to build an importance table which allows attributes to be ranked ordinally and by relative magnitude. A case study utilizing this method is discussed and its results are presented and summarized.

Share

COinS