CSCE 633
Machine Learning (Spring 2010)
due: Tues, Feb 23

Project 1
---------

Goal: Implement and test a decision tree algorithm (ID3).  You should
also implement reduced-error pruning.

You may use any programming language you like.  However, the code must
be own, not shared with another student or copied from a source on the 
web.

You will have to implement a simple parser to read in datasets in the
format provided by the UCI Repository (typically column-separated
values, one example per row).  You will probably want to implement
a simple routine for analyzing attribute values and summarizing their
distribution.

At your discretion, you may experiment with: alternative splitting
criteria, stopping criteria, or pruning methods.  You might want to
compare binary with multi-way splits.  You might want see if there
are cases where using GainRatio has an advantage.  You might want
to test different methods for handling missing values.  The choice
of what to do is up to you - just make it interesting.

What to turn in: 

A written project report (in the format of a conference paper) that
describes significant details of your implementation, results of any
comparative tests you do, and performance (accuracy) on several
databases.

Choose several (at least 5) datasets from the UCI ML Repository,
including at least one with all numeric variables (like Iris) and one
with all categorical variables (like Congressional Voting).  At
minimum, you should test your DT algorithm against a simple majority
classifier, with pruning, and without pruning.  You should use proper
statistics to report performance, such as cross validation, confidence
intervals on accuracy, and paired t-test for comparisons to support
any conclusions you draw.