ANU Data Mining Group

News:

• 11 October: Added slides for fifth predictive modelling lecture; added assignment for module 4.

• 6 October: Added all student presentations (see below) and included proposed dates and times for presentations.

• 5 October: Uploaded lecture slides for third and fourth predictive modelling lectures - see below.

Lecture and lab times and venues

• All lectures are being held in the John Dedman Mathematical Sciences Building [no. 27], room G35.

  Lecture times:

  1. Tuesday 11-13
  2. Wednesday 12-13

  Please note that not all lecture slots will be used.

• Venue for labs: John Dedman Mathematical Sciences Building [no. 27], room LG 15 or 16 (computer lab downstairs).

  Lab time:

  1. Wednesday 10-12

  Please note that labs will start in week 2.

• Course schedule:
  • Weeks 1 and 2: Module 1 (Peter Christen)
  • Weeks 3 and 4: Module 2 (Graham Williams)
  • Weeks 5, 6 and 7: Module 3 (Markus Hegland)
  • Week 8: Mike Osborne's Workshop (Lectures and lab cancelled)
  • Weeks 9, 10 and 11: Module 4 (Stephen Roberts and Jochen Garcke)
  • Weeks 12 and 13: Student presentations

Course coordinator Stephen Roberts

Prerequisits
We don't actually have formal prerequisites. We assume that participants have basic computing and mathematical skills with honours level expertise in either of these areas. Our intended audience are honours and graduate students in these areas. We intend to look at both mathematical (algorithm) and computing issues.

In the course we will analyse algorithms, do some coding and look at efficiency issues, and also do some hands on data analysis (data mining).

Lecturers

• Peter Christen
  ANU Department of Computer Science

• Jochen Garcke
  ANU Mathematical Sciences Institute, Centre for Mathematics and its Applications

• Markus Hegland
  ANU Mathematical Sciences Institute, Centre for Mathematics and its Applications

• John Maindonald
  ANU Mathematical Sciences Institute, Centre for Bioinformation Science (CBiS)

• Stephen Roberts
  ANU Mathematical Sciences Institute

• Graham Williams
  Togaware / ATO

Course Assessments

The assessment for this year's course is as follows:

1. Four assignments (one per course module) each worth 10%
2. R programming hands-on worth 20%
3. Paper presentation worth 20% (in weeks 12 and 13)
4. Examination (take home or written) worth 20%
5. The final course mark will be the sum of the assignments, presentation and exam marks.

Course Outline (draft)

0. Welcome and introduction
   (Peter Christen)
   Lecture slides (4-up): Lect-0, PDF / Lect-0, PS.GZ

1. Data mining overview, process, data issues, data quality, data preprocessing, data integration / linkage
   (Peter Christen)
   Lecture slides (4-up): Lect-1, PDF / Lect-1, PS.GZ     Lect-2, PDF / Lect-2, PS.GZ     Lect-3, PDF / Lect-3, PS.GZ     Lect-4, PDF / Lect-4, PS.GZ

2. Introduction to data mining algorithms (from a machine learning point of view, incl. clustering)
   (Graham Williams)
   Lecture slides: Lect-5, PDF, 4-up     Lect-6, PDF, 4-up     Lect-7, PDF, 4-up     Lect-8, PDF, 4-up     Lect-9, PDF, 4-up     Lect-10, PDF, 4-up

3. Algorithms for association rules
   (Markus Hegland)
   Lectures 1, PDF

4. Predictive modelling
   (Stephen Roberts and Jochen Garcke)
   Lect-1 PDF, 4-up     Lect-2 PDF, 4-up     Lect-3 PDF, 4-up     Lect-4 PDF, 4-up     Lect-5 PDF, 4-up

Assignments

• The assignment for module 1 is due on Friday 12 August, 5 pm.

• The assignment for module 2 is due on Friday 2 September.

• The assignment for module 3 is due on Tuesday 27 September 2005.

• R assignment is now available on John's Web site as PDF file. Due date is October 19.

• The assignment for module 4 is due on Wednesday 16 November (files for graph: svmData.eps, svmData.png).

• Student presentations:
  Each student is required to select a data mining paper from the ACM KDD 2005 research track. Please e-mail the title and authors of your selected paper to Steve Roberts.
  - The presentations will be in weeks 12 and 13 (detailed schedule will be announced later).
  - Presentations should be 20 minutes long. Further information will be given in the lectures.

  Selected talks are (so far):

  Sarah Dexter (Tuesday 18th Oct 11:00)	Determining the authors native language by mining a text for errors by Mosche Koppel, Jonathan Schler and Kfir Zigdon.
  Lewis Conn (Tuesday 18th Oct 11:30)	Learning to Predict Train Wheel Failures by Chunsheng Yang and Sylvain Letourneau.
  Sarah Bolt (Tuesday 18th Oct 12:00)	Mining risk patterns in medical data by Jiuyong Li et.al.
  Peter Humburg (Tuesday 18th Oct 12:30)	Simple and effective visual models for gene expression cancer diagnostics by Gregor Leban et.al.
  Song Yang Guo (Wednesday 19th Oct 12:00)	Data mining in the chemical industry by Alex Kalos and, Tim Rey
  Eleanor Donovan (Wednesday 19th Oct 12:30)	Deriving marketing intelligence from online discussion by Natalie Glance et.al.
  Michael Kehoe (Tuesday 25th Oct 11:00)	Formulating distance functions via the kernel trick by Gang Wu, et.al.
  Kurt Pudniks (Tuesday 25th Oct 11:30)	Automated detection of frontal systems from numerical model-generated data by Xiang Li, et.al.
  Thomas Sutton (Tuesday 25th Oct 12:00)	Variable latent semantic indexing by Anirban Dasgupta et.al.

Labs / Tutorials

• Statistical perspective on data mining
  (John Maindonald and Kevin Wang)
  Hands-on statistical data mining exercises based on R.

• An overview of the lab sessions is available from John's Web page.

• Tutorial 1 handout   Tutorial 1 solutions

• Tutorial 2 handout   Tutorial 2 solutions

• Tutorial 3 handout   Tutorial 3 solutions