ECE 8820 Pattern Recognition

Computer Project #2

Fall, 2008

Due November 18 

Part I: Consider the (crisp) K-Nearest Neighbor Algorithm discussed in class.

Using the 2 class, 4 dimensional dataset from project #1, experiment with different values of K and report the classification rates in a confusion matrix. Test the algorithm using features 1 and 2 only and using features 3 and 4 only. (That is, you do NOT have to use all 4 features together.) To simplify the testing, reserve the first 10 % of each class data for testing and use the remaining data as the training set. (If you are feeling ambitious, you can test using 10 fold cross validation; as before, the first fold uses the first 10% of each class data for testing and the remaining data as the training set.) Report the confusion matrix for each test.

Part II: Now consider the Fuzzy K-Nearest Neighbor Algorithm discussed in class. Again using the 2 class, 4 dimensional dataset, repeat part I above using the Fuzzy K-NN algorithm. Experiment with different values of K. Try using crisp memberships of the training data and also setting the fuzzy memberships of the training data as we discussed in class. Report the confusion matrix for each test.

As before, your report should contain a sections on

1. The technical description of all techniques utilized,
2. The design of the algorithms (pseudo-code, flowcharts, or some other structured descriptive means),
3. The results of the algorithms (Describe your experimental conditions with the results so it is clear what experiment was done for each set of results.)
4. An analysis of the results, i.e., did you obtain what you expected? Were there any surprises? What were the challenges of the project? What conclusions can you draw from the experiments? etc. Your analysis should include a comparison of the results using the crisp and fuzzy K-NN algorithms and also a comparison with the Bayes minimum error classifier from the first project.
5. Well documented, structured, modular program listings.