1. Looking at the dataset: Below we use 3 alternate ways of looking at a dataset:

   1. Using Weka's ArffViewer:

      1. Run the Weka system.

      2. Select "Tools" and then "ArffViewer".

      3. Go to "File" → "Open" → "data" → "iris.arff" and open this file.

   2. Using a word editor to access the dataset file directly:

      1. Separately, use a word editor (e.g., WordPad, emacs, ...) to open the file where the Iris dataset is stored. If WekaDirectory were the name of the directory where the Weka files reside, then the Iris dataset would be stored at: WekaDirectory/data/iris.arff

      2. * Include the header of this dataset in your report (just the header, NOT the data).

      3. Compare this file with the display of this dataset in the ArffViewer above.

   3. Using Weka's Explorer:

      1. From the main Weka window, select "Applications" and then "Explorer".

      2. Go to "Preprocess" → "Open file" → "data" → "iris.arff" and open this file.

      3. * Record in your report the minimum and maximum value for each numeric attribute in the dataset (you can find this info in the Explorer display).

2. Analyzing the raw dataset

   1. From the Weka's Explorer window, select "Classify".

   2. "Choose ZeroR" and click on "Start".

   3. * Record in your report:
      1. The class predicted by ZeroR. Also, provide a clear and concise explanation of why that's the class chosen by ZeroR.
      2. The accuracy of this ZeroR prediction (that's is the percentage of correctly classified instances).
      3. The confusion matrix. Give a brief description of why this confusion matrix contains non-zero values in just one column.

   4. Go back to the "Classify" window of Weka's Explorer. Click on "Choose", select "OneR" and click on "Start".

   5. * Record in your report:
      1. The rule output by OneR.
      2. The accuracy of OneR's prediction (that's is the percentage of correctly classified instances).
      3. The confusion matrix of OneR's prediction.
      4. Select and Right-click on "rules.OneR" under the "Result list" part of the Classify window. Select "Visualize classifier errors". Play with different selections of "X" and "Y" axes for the display until you find a display that you can understand and explain. Include this display together with your explanation of it in your report.

3. Preprocessing the dataset before analysing it.

   1. Discretize the numeric attributes in this dataset:

      1. On the Preprocess tab of the Weka's Explorer window, go to "Filter" and click "Choose". Select "filters" → "supervised" → "attribute" → "Discretize".

      2. Back on the Filter part of the Preprocess tab of the Weka's Explorer window, click on "Apply".

      3. * Using the attributes' information displayed on the Explorer window, include in your report the values of each of the attributes in the filtered dataset and how many data instances contain each value.

   2. * Using this discretized dataset, run ZeroR again. Record in your report the class predicted by ZeroR, the accuracy of this prediction, and the confusion matrix. Are there any differences between the results for ZeroR over the discretized data and over the raw data? Explain why there are or there are no differences.

   3. * Using this discretized dataset, run OneR again. Record in your report the class predicted by OneR, the accuracy of this prediction, and the confusion matrix. Are there any differences between the results for OneR over the discretized data and over the raw data? Explain why there are or there are no differences.

4. Weka Code. * Find the Java code that implements the discretization filter in the directories that contain the Weka files. Include the code implementing this filter in your report, and describe the algorithm followed by this code in your own words.

   You can find the Weka code in a file called "weka-src.jar", which should be located in the directory where Weka was installed. This "weka-src.jar" file is a zip file. Hence you need to winzip or unzip it to extract its contents. Inside, you will find the .java files that implement Weka.

The census-income dataset from the US Census Bureau which is available at the Univ. of California Irvine (UCI) Data Repository.
The census-income dataset contains census information for 48,842 people. It has 14 attributes for each person (age, workclass, fnlwgt, education, education-num, marital-status, occupation, relationship, race, sex, capital-gain, capital-loss, hours-per-week, and native-country) and a boolean attribute class classifying the input of the person as belonging to one of two categories >50K, <=50K.

1. Convert the census-income data to the arff format. For this you can either use any tools provided by Weka, or you can make the conversion outside the Weka system using other tools (e.g., a word editor, Excel, etc.). Create a census-income.arff file with the converted dataset.

2. * Include in your report the header of your census-income.arff file together with the 10 first data instances of the dataset. (Do NOT include the full dataset - Just the first 10 data instances.)

3. Load this dataset into Weka by opening your census-income.arff dataset from the "Explorer" window in Weka. The full dataset should fit in main memory, but if it doesn't, remove as many instances as you need (but not more!) from your census-income.arff file (do this outside Weka using for instance a word editor or Excel).

4. * Run the "ReplaceMissingValues" filter (available at Filters → Unsupervised → Attribute) over this dataset. Explain clearly in your report what this filter does to the dataset.

5. * Run the "Resample" filter (available at Filters → Unsupervised → instance) This filter selects a random sample from the data instances in the current dataset.
   1. Click on "More" to learn more about this filter
   2. Use say "20.0" in the "SampleSizePercent" to select a subsample of 20% of the current instances.
   3. Click "OK"
   4. Click "Apply" on the main Weka Window
   Describe in your report the meaning of each of the parameters of this Resample filter.