CSI 991

Data Mining

Fridays 3:00pm -- 5:00pm, Innovation Hall, Room 209

The course is CSI 991, Section 004.

Contacts:
csutton@gmu.edu
jgentle@gmu.edu

• September 12
  Introductory discussions.

• September 19
  Seminar canceled because of storm.

• September 26
  
  • Material from Chapters 1 and 2; presentation by Yasmin Said.
  • Binning and other topics by Dan Carr.

• October 3
  Questions/disscussions for Chapters 2 through 4.

• October 10
  
  • CART and related classification method by Cliff Sutton. (Note: The corner of p. 23 has a mistake --- it should be 3*99 = 297 instead of (99)^3.)
  • Perceptrons, neural networks, and data mining by Jim Shine.
  • The a priori algorithm by Daniel Barbara.
  • More on Chapter 5.

• October 17
  
  • Datamining text and graphics on the web by Phil Sage.
  • First four sections of Chapter 6.
  • Mutual information schematic by Daniel Barbara.

• October 24
  
  • Multi-dimensional range queries and nearest neighbor classification, by Carlotta Domeniconi (related paper: An efficient density-based approach to data mining tasks, by Carlotta Domeniconi and Dimitrios Gunopulos.)
  • Some clarification on projection pursuit (not on projection pursuit regression) by Jim Gentle.
  • Projection pursuit regression and neural nets by Jim Shine.
  • Normal mixtures and density estimation.
  • Other topics in the remainder of Chapter 6 (Section 6.6) and in Chapter 7.

• October 31
  
  • Hidden Markov models by David DeBarr.
  • Cross validation and more on CART by Cliff Sutton (see material on October 10)

• November 7
  Chapter 8; general discussion.

• November 14
  First, we consider some intro material on the EM algorithm (Gentle)
  Then we experiment with classification methods on the county election results data, which is a modified (and tab-delimited!!!) dataset from Harrel's book.
  The following changes were made to Harrel's original dataset:
  • Deleted the 27 counties that were missing the dependent variable
  • Derived pdensity (log of population density)
  • Derived senior (sum of age6574 and age75)
  • Derived clinton (binary indicator set to 1 if democrat > republican and democrat > perot; else set to zero)
  • renamed pop.density popdens
  • renamed pop.change popchan
  • renamed republican republic
  Jill showed how SPSS Clementine works on this dataset.
  

• November 21
  Presentation on cluster analysis by Pragyansmita Nayak
  Continuation of the study of the "counties" data, using CART and S-Plus.

• November 28
  No seminar; Happy Thanksgiving!

• December 5
  Chapter 10.

• December 12
  Chapter 11.