Initial Experiment Results

      All experiments here were done with the full training data set and tested with the full test data set. For the details of each experiment, please see the test.txt file for that test.

      The best results from this set of testing was the Kernel Density Estimator test (uses one Gaussian kernel per observed data value) instead of a normal distribution. Since this test was run out of my own curiosity for the complex Naive Bayes Classifier provided with Weka, I am not including it with the other test results using a simple Naive Bayes Classifier which more closely resembles the C code provided by the book. I provide the results here for completeness. The overall success rate for this classifier was 83.0047%.

      The next best results were obtained from Default settings. This used all the training data and all the testing data to achieve an 82.9924% success rate. It is interesting to note that when training examples with missing attribute values were removed from the training set, the success rate went down to 82.4104%.

Summary of these results:

      Based on the discussion in class (October 19, 2000) I decided to try two data pre-processing approaches.

1. normalizing the continuous attributes
2. discretizing the continuous attributes