Weka Machine Learning Framework Implementation Report

1. Data Loading and Preprocessing

The project leverages Weka's data loading and preprocessing capabilities through several key components:

Data Loading Strategy

• CSVLoader: Used in DataLoader class to convert CSV files to Weka's native ARFF format
• ArffLoader: Used in ClassificationFramework to load processed ARFF files
• Conversion Process:
  • CSV files are first loaded using CSVLoader
  • Data is then preprocessed (removing unnecessary attributes, setting class index)
  • Data can be saved as ARFF using ArffSaver

Data Preprocessing Techniques

• Standardization: Uses Weka's Standardize filter to normalize numeric attributes

  Standardize standardize = new Standardize();
  standardize.setInputFormat(fullData);
  fullData = Filter.useFilter(fullData, standardize);

• Automatic attribute removal
• Class index setting
• Optional data sampling for large datasets

2. Prediction Framework

Classifier Abstraction

• Interfaces and Abstract Classes:
  • IModel interface defines core method signatures
  • BaseClassifier provides common implementation for classifier methods
  • Specific classifier classes (LinearRegression, SVMRegression, etc.) extend BaseClassifier

Model Training and Evaluation

• Supports multiple regression models:
  • Linear Regression
  • SVM Regression
  • M5P Decision Tree
  • Random Forest

Evaluation Methods

1. Train-Test Split:

   • 80% training, 20% testing data
   • Manual prediction and error calculation

2. Cross-Validation:

   • 10-fold cross-validation using Weka's Evaluation class
   • Calculates metrics like:
     • Correlation coefficient
     • Mean absolute error
     • Root mean squared error
     • Relative error percentages

3. Model Configuration

Each classifier is configured with specific Weka API parameters:

Example: SVM Regression Configuration

SMOreg svm = new SMOreg();
svm.setC(1.0);  // Complexity parameter
svm.setFilterType(new SelectedTag(SMOreg.FILTER_NORMALIZE, SMOreg.TAGS_FILTER));

PolyKernel polyKernel = new PolyKernel();
polyKernel.setExponent(1.0);
svm.setKernel(polyKernel);

Example: Linear Regression Configuration

lr.setAttributeSelectionMethod(new SelectedTag(1, LinearRegression.TAGS_SELECTION));
lr.setRidge(1.0E-8);
lr.setEliminateColinearAttributes(true);

4. User Interface Integration

• GUI allows interactive:
  • Dataset loading
  • Classifier selection
  • Model training and evaluation
• Background threading for model training
• Real-time logging of training process

Key Weka API Highlights

• Instances: Core data structure
• Filter: Data preprocessing
• Classifier: Model training
• Evaluation: Performance metrics
• Kernel and optimization configurations

Conclusion

This implementation demonstrates a comprehensive machine learning workflow using Weka's powerful APIs, providing flexibility in data preprocessing, model training, and evaluation.