Data Preprocessing in WEKA

Step 1: Open Weka

• Launch WEKA GUI Chooser.
• Click on "Explorer" to open the main interface.

Example File : weather.arff

This is a small dataset about weather conditions and whether to play tennis.

Step 2: Load a Dataset

• Click "Open file" in the Preprocess tab.
• Navigate to: C:\Program Files\Weka-X.X\data\weather.arff (or wherever WEKA is installed).
• Select and open weather.arff.

Step 3: Understand the Dataset

• Once loaded, you can see:
  • Attributes (e.g., Outlook, Temperature, Humidity, Windy, Play)
  • Instances (e.g., 14 rows)
• Click on any attribute to see its value distribution.

Step 4: Remove Unwanted Attributes

• Select the attribute (e.g., "humidity").
• Click Remove if you don't want to include it in your analysis.

Step 5: Handle Missing Values

• If your dataset has missing values:
  • Click "Choose" under the Filter section.
  • Go to: filters > unsupervised > attribute > ReplaceMissingValues
  • Click Apply to replace all missing values.

Step 6: Normalize or Standardize Data (if needed)

• For numeric data:
  • Use Normalize or Standardize filters:
    • filters > unsupervised > attribute > Normalize
    • filters > unsupervised > attribute > Standardize
  • Click Apply.

Step 7: Convert Nominal to Numeric (if required)

• Some classifiers (like Neural Networks) require numeric input.
• Use: filters > unsupervised > attribute > NominalToBinary or NominalToNumeric

Step 8: Discretization (convert numeric to nominal)

• Useful for algorithms like ID3.
• Use: filters > unsupervised > attribute > Discretize

Step 9: Save the Preprocessed Data

• Click Save to export the modified dataset.
• Choose format: .arff or .csv

Example Summary (weather.arff)

• Remove attribute: Humidity
• Apply filter: ReplaceMissingValues
• Normalize temperature
• Save as: weather_processed.arff

Notes

• Filters can be unsupervised (independent of class) or supervised (depend on class attribute).
• Always analyze attributes visually to detect issues (e.g., outliers, missing values).