Machine Learning

Homework 5

Due: October 29, 2010 (midnight)

For this assignment you will become familiar with the SimpleLogistic and MultilayerPerceptron learning algorithms.

  1. Using WEKA, execute the SimpleLogistic classifier using default parameters on the Iris dataset using the training set as the test option.
    1. Show the output of WEKA.
    2. Describe the meaning of the classifier model.
  2. Using WEKA, execute the MultilayerPerceptron classifier using default parameters on the Iris dataset using the training set as the test option.
    1. Show the output of WEKA.
    2. Draw a picture of the network, including input nodes and their associated features, hidden layer nodes, and outputs nodes and their associated classes. Also include the links, but no need to include the actual weights. Be sure to include the thresholds.
    3. Describe in your own words the meaning of the default parameters: -L, -M, -N and -H. Also include the likely effect on learning when these parameters are changed.
    4. Determine a setting for these four parameters that leads to zero error on the training set.
  3. Next we will compare the MultilayerPerceptron and J48 classifiers.
    1. Perform a 10-fold cross-validation experiment comparing MultilayerPerceptron and J48 using default parameters on the four datasets: diabetes, iris, labor, and vote.
    2. Show a table with the accuracy rates for each dataset and classifier. Indicate with an asterisk "*" any dataset where one classifier significantly outperforms the other with 95% confidence.
  4. Compare the performance of the MultilayerPerceptron and J48 classifiers based on the results from the previous problem. Specifically, which classifier performs better on which datasets and why. The "why" part should consider the characteristics of the data, the hypothesis space, and the learning algorithm.
  5. Email to me (holder@eecs.wsu.edu) your nicely-formatted report (PDF preferred) containing your responses to the above problems.