On relationships among various cost functions and with an uncertainty measure for decision trees

Speaker(s)

Shahid Hussain

Affiliation

KAUST

Date

June 17, 2011, 2:15 p.m.

Room

room 5820

Seminar

Research Seminar of the Logic Group: Approximate reasoning in data mining

---

This talk is devoted to the design of new tools for studying exact and approximate decision trees (α-decision trees). We present algorithms to study the relationships between number of misclassification and number of nodes with the depth of an exact decision tree. We also present an algorithm that describes relationships between these and other cost functions (such as average depth, number of terminal nodes, etc) with a measure of uncertainty for approximate decision tree α -decision trees). Furthermore, we provide results of experiments conducted with these algorithms on decision tables (datasets) acquired from UCI ML Repository.

---