Data Mining Using Recursive Partitioning Peter Westfall With some help from Dr. Barry Macy, Dr....

14
Data Mining Using Recursive Partitioning Peter Westfall With some help from Dr. Barry Macy, Dr. Seul-Hee Yoo, and TTU Institutional Research

Transcript of Data Mining Using Recursive Partitioning Peter Westfall With some help from Dr. Barry Macy, Dr....

Page 1: Data Mining Using Recursive Partitioning Peter Westfall With some help from Dr. Barry Macy, Dr. Seul-Hee Yoo, and TTU Institutional Research.

Data Mining Using Recursive Partitioning

Peter WestfallWith some help from

Dr. Barry Macy, Dr. Seul-Hee Yoo, and TTU Institutional Research

Page 2: Data Mining Using Recursive Partitioning Peter Westfall With some help from Dr. Barry Macy, Dr. Seul-Hee Yoo, and TTU Institutional Research.

Business Intelligence

= Transforming Business Data into Action

What Data?Lots of data.http://www.pcworld.com/news/article/0,aid,113170,00.asp

Text, numeric, sound, pictures, video.

Page 3: Data Mining Using Recursive Partitioning Peter Westfall With some help from Dr. Barry Macy, Dr. Seul-Hee Yoo, and TTU Institutional Research.

Old and New Learning Paradigms

Old:

THEORYData

Analysis THEORY

New:Theory

DATAANALYSIS

DATAANALYSIS

Page 4: Data Mining Using Recursive Partitioning Peter Westfall With some help from Dr. Barry Macy, Dr. Seul-Hee Yoo, and TTU Institutional Research.

Typical Data Mining Methods

• Clustering (eg, customer segmentation)

• Affinity (eg, what items do people buy together)

• Exception analysis (eg, credit card fraud, terrorism)

• Predictive Modeling (eg, deciding loans, predicting employee turnover, predicting likely customers)

Page 5: Data Mining Using Recursive Partitioning Peter Westfall With some help from Dr. Barry Macy, Dr. Seul-Hee Yoo, and TTU Institutional Research.

Recent Horizons in Data Mining

• Visualizations

• Text mining

• Audio mining

• Video mining

Page 6: Data Mining Using Recursive Partitioning Peter Westfall With some help from Dr. Barry Macy, Dr. Seul-Hee Yoo, and TTU Institutional Research.

Requirements of DM Tools

• Simple (even an MBA can use it)

• Actionable results

• Flexible, open-ended (“Analysis at the speed of thought”)

• Scale-Up: Can handle massive data sets

• Drill-Down: Ability to investigate sub-units

Page 7: Data Mining Using Recursive Partitioning Peter Westfall With some help from Dr. Barry Macy, Dr. Seul-Hee Yoo, and TTU Institutional Research.

Recursive Partitioning

• A predictive modeling tool• Also called “Decision Trees”, “CART”• Works by recursively splitting data set• Software:

– SAS Enterprise Miner– SPSS Clementine– SPLUS– Lots of Freeware– Demo: “Partitionator” of Eureka! Technologies.http://www.eurekatechnologies.com/MoreDetails.aspx

Page 8: Data Mining Using Recursive Partitioning Peter Westfall With some help from Dr. Barry Macy, Dr. Seul-Hee Yoo, and TTU Institutional Research.

Example 1: Survey of Innovative Organizations

• Action Orientation: Which management levers lead to better performance?

• V24=earned profit in last 5 years: – 1=all five – 2=most of 5 – 3 = some of five – 4 = none of 5

Page 9: Data Mining Using Recursive Partitioning Peter Westfall With some help from Dr. Barry Macy, Dr. Seul-Hee Yoo, and TTU Institutional Research.

Interesting Variables

• V617B = Number of years that elimination of perks for certain groups of people has been in effect

• V894A = Percent of workforce involved in SPC/SQC/TQC training– 1=None – 2=1-20%– …– 7=100%

Page 10: Data Mining Using Recursive Partitioning Peter Westfall With some help from Dr. Barry Macy, Dr. Seul-Hee Yoo, and TTU Institutional Research.

Example 2:Texas Tech University Ratings

By Thesis Students

• Who is satisfied? Who is not satisfied?

• Action Orientation – – Improve pockets where students are

dissatisfied.– Emulate pockets where students are satisfied.

Page 11: Data Mining Using Recursive Partitioning Peter Westfall With some help from Dr. Barry Macy, Dr. Seul-Hee Yoo, and TTU Institutional Research.

Example 3: Business Dress Styles Rated

Page 12: Data Mining Using Recursive Partitioning Peter Westfall With some help from Dr. Barry Macy, Dr. Seul-Hee Yoo, and TTU Institutional Research.

Lower Rated Dress Types

Page 13: Data Mining Using Recursive Partitioning Peter Westfall With some help from Dr. Barry Macy, Dr. Seul-Hee Yoo, and TTU Institutional Research.

Final Tree – Dress Ratings

Page 14: Data Mining Using Recursive Partitioning Peter Westfall With some help from Dr. Barry Macy, Dr. Seul-Hee Yoo, and TTU Institutional Research.

Questions?

Comments?

Poison-tipped darts?