Project 2005

1
Project 2005

• F2E5414 Pattern classification

2
Procedure

• Choose problem
• Either an artificial three category
  classification problem.
• Your own classification problem (and your own
  data)
• Or something completely different
• Choose methods
• One advanced method
• Gaussian mixture model based Bayes classifier
  (GMM)
• Neural network classifier (NN)
• Support vector machine (SVM)
• Three benchmark methods
• One from each of the chapters 3, 4, and 5
• Implement, simulate, write report, and give a
  short presentation
• See the pdf-document on web

3
Artificial Problem

• Three categories
• Uniform class pdf
• Gaussian class pdf
• Laplacian class pdf
• Equal priors
• Two-dimensional feature space
• Minimize error rate

4
Tasks

• Calculate optimal Bayes classifier decision
  boundaries
• Calculate minimum errror rate.
• For each of the four chosen classifiers
• Whenever possible, provide analytical expression
  of the decision boundaries
• Provide a graphical illustration of the decision
  boundaries
• Study how the amount of training data affects the
  error rate
• For GMMs and NNs
• For a fixed database size, study how the number
  of components/hidden units affects the error rate

5
Requirements

• Perform all tasks
• Write a short, concise report
• Give a 15 minute presentation
• Successful completion of the above yields 3
  credits.