CSCE 633
Machine Learning (Spring 2010)
due: Tues, Apr 6

Project 3
---------

Goal: Implement and test a nearest-neighbor learner, and compare
its performance to your decision-tree and neural network.

Although a nearest-neighbor learner is simple in principle, there are
a number of design issues you might want to test. For example, which
is best: 1-NN, 3-NN, 5-NN, or using distance-weighting?  Also, you
might consider using the noise-tolerant editting strategy we talked
about in class (see the NTGrowth algorithm), where you keep only those
training examples on which mistakes are made based on previous
examples.  Or perhaps try identifying representative/prototypical
instances to save using clustering.  Another suggestion is that you
might consider implementing and testing a simple form of feature
selection or feature weighting.