DATA MINING SPECIAL TOPICS CLASS

• Time: Thursday, 11:30-2:30
• Location: Hill Center 552
• Instructor: David Madigan
• Email: madigan@stat.rutgers.edu

• Course Syllabus

  Here's a brief description of what this course is about.

• Course Text

  We will be using a draft version of Principles of Data Mining , by Hand, Mannila, and Smyth (MIT Press, forthcoming), as the course text.  Details on how to get copies of the draft text will be announced in class.

  We will also make use of J. Han and M. Kamber (2000) Data mining: concepts and techniques. Morgan Kaufman. Han and Kamber have a terrific set of lecture notes on the Web.

• Projects

  The class will feature a project component and I will use the projects to assign a grade. Students should come up with a project proposal by the end of February and provide me with a one-page proposal. In-class project presentations will take place in the last couple of weeks of the semester. Projects should start with a meaningful question-of-interest and a target dataset of at least moderate size. There should be an evaluation component to the work - how well do the methods you use address the question you started out with?

Class Topics.
I will post links to materials we use in the class here.

DATE	TOPICS	LINKS
January 18	Introduction to Data Mining HTML A warehousing and mining case study Association Rules	This is an edited version of Han and Kamber's Chapter 1 notes - See Han and Kamber's Chapter 6 notes
January 25	Relational databases and SQL Data Warehousing Ideas for projects HTML Visualizing and Exploring Data HTML PPT	There are excellent notes here and here. See Han and Kamber's Chapter 2 notes - The OLIVE Library
February 1	Some basic principles HTML PPT Ideas for projects HTML *Support Vector Machines HTML PPT	Chapter 4, 5, & 6 of HMS - -
February 8	Descriptive modeling HTML PPT Graphical Markov Models-Intro Ideas for projects HTML *Global Partial Orders from Sequential Data	Chapter 9 of HMS - - PDF PS
February 15	Graphical Markov Models-Gaussian UDGs Ideas for projects HTML Score Functions HTML PPT	- - Chapter 7 of HMS
February 22	Predictive Models for Classification HTML PPT Bias versus Variance for Classification	Chapter 10 of HMS PS PDF
March 1	Bias correction for Naive Bayes Search and Optimization Methods HTML PPT *Flexible Metric NN Classification	- Chapter 8 of HMS ps.Z PPT
March 8	*Variational methods Monte Carlo methods	- -
March 15	Spring break	-
March 22	*Likelihood-based squashing Bayesian networks	PDF -
March 29	Bayesian networks continued	-
April 5	Bagging, Boosting, etc. PPT Neural Networks	Check this list. See also Greg Ridgeway's Interface paper. PPT (slides by Sebastian Thrun)
April 12	Information retrieval and filtering	-
April 19	I need to reschedule this class	-
April 26	Project presentations	-

* indicates a "special topic"

Software for Projects.
You can use any software you like for the projects. I recommend that you use tools like Matlab, SAS, or S-Plus so as not to get too bogged down in software issues. Feel free to either use available implementations of data mining algorithms (many are publically available, e.g., see the UC Irvine Machine Learning Data Archive for a list of data mining/machine-learning software or you can implement an algorithm directly in a language of their choice (naturally you will learn more if you opt for the 2nd option, but please make sure you understand the complexity of the project you are proposing and that you will have time to finish it). Check the kdnuggets site for an extensive listing of software tools.

Other Data Mining Courses

• Padhraic Smyth's course at UCI

General Pointers to Data Mining and related Web pages

• The ACM SIGKDD is the premier professional organization for data mining.
• Proceedings of KDD-2000
• Knowledge Disovery and Data Mining. The main Web page for KDD related information.
• STATLIB. STATLIB is a major online source of information for statistical computing.
• Machine Learning Resources. Very extensive list of information and pointers on machine learning research and applications.
• Pattern Recognition. Information and resources about pattern recognition methods and applications.

General Benchmark Data Collections

• The UC Irvine KDD Archive  This is a new archive of large data sets specifically collected for researchers in data mining and KDD to serve as a benchmark collection of data (developed under sponsorship of the National Science Foundation). There are some excellent data sets here to consider as class projects.
• STATLIB data sets. STATLIB is the major online archive for statistical research and the data sets portion contains many data sets which have been analyzed in the statistical literature (mostly smaller data sets, many for regression, classification, and other statistical methods).
• UC Irvine Machine Learning Data Archive. The ICS department's widely known original archive of benchmark data sets (mostly smallish data sets, mostly for classification problems). I would prefer you look first at the larger data sets from the UCI KDD archive (above) for class projects. Nonetheless, there are some interesting data sets here too.
• Computer Vision Test Images. Very extensive set of pointers to online collections of sets of image data.
• The DELVE benchmark collection. A collection of data sets and software focused on the systematic evaluation of learning algorithms.
• A Casebook for a First Course in Statistic and Data Analysis. A very nice collection of "smallish" data sets based on a text of the same name, including many classic data sets such as Iris and Old Faithful.
• Time Series Data Sets. Very extensive collection of over 500 time series data sets from a very broad collection of applications.
• DASL: The Data and Story Library. Very nice collection of "small" data sets with background descriptions, used mainly for teaching.
• NIST Benchmark Image Data Sets. Benchmark data sets for optical character recognition and face recognition problems.
• CHANCE Data Sets. Small collection of data sets used in the CHANCE teaching framework (including Dow Jones and some sports-related data).

Collections of Application-Specific Data Sets

• Economic Time Series Data. Contains virtually every economic time series (for the United States) that you can imagine.
• Asteroid Elements Database A dataset containing detailed information on several thousand asteroids in the solar system. The problem of interest here is one of clustering or grouping the asteroids into natural groups.
• Climate Data Sources. Very extensive list of pointers to climate, atmosphere, and ocean related data sets.
• Comet Library. Large collection of images of observed comets.
• El Nino and Internet usage data. (Not related!). Two nicely documented data sets intended as benchmarks for the 1999 American Statistical Association meeting.

Sites which have lists of various data set collections

• Knowledge Discovery and Data Mining Data Sets. Pointers to data sets used in KDD research.
• Pattern Recognition Archives. Pointers to data archives used in the pattern recognition, machine learning, and neural network research literature.
• University of Nevada at Reno. Extensive list of pointers to various online data archives.
• UCLA Statistics Department. Pointers to data sets from various texts and national studies and collections.