Data Visualization in WEKA

Step 1: Launch WEKA

• Open WEKA GUI Chooser.
• Click on Explorer to launch the main interface.

Step 2: Load a Dataset

• In the Preprocess tab:
  • Click Open file.
  • Load the dataset weather.arff (from the data folder inside the WEKA installation directory).

Step 3: View Attribute Information

• After loading:
  • You'll see all attributes listed on the left.
  • Select any attribute (e.g., "outlook").
  • On the right, WEKA displays:
    • Histogram for nominal attributes.
    • Box plot or line plot for numeric attributes.
  • This shows the distribution of values.

Step 4: Visualize All Attributes

• Click on the Visualize All button (bottom-right corner).
• This opens a scatter plot matrix, showing relationships between all attributes.
• Each square shows how two attributes are related.
• Color is based on the class attribute (e.g., play=yes/no).

Step 5: Customize Visualization

• In the Visualize All window:
  • You can change:
    • X-axis and Y-axis attributes using drop-down menus.
    • Coloring attribute (usually the class).
  • You can also zoom in, drag, or select points.

Step 6: Visualize Individual Plots

• Use the Visualize tab (after classification or clustering):
  • After running a model, go to the "Visualize classifier errors" to see prediction distributions.
  • View ROC curves, error plots, and more.

Step 7: Export Visualizations

• Right-click on the chart area.
• Select Save or Copy to export the visualization as an image.

Example Use Case

Using weather.arff:

1. Select the attribute "temperature".
2. WEKA shows a histogram of different temperature values.
3. Click Visualize All → You can see how temperature and humidity relate.
4. Color the graph using class "play".
5. Observe patterns (e.g., high humidity tends to "no" play).

Tips:

• Use filters (like discretize or normalize) to prepare data for better visualization.
• After running classification, use the "Visualize classifier errors" to understand how your model behaves.
• You can use "Attribute Selection" to reduce clutter in plots.