Pruning Decision Rules by Reduct-Based Weighting and Ranking of Features

Area of Science:

• Machine Learning
• Data Mining
• Artificial Intelligence

Background:

• Feature selection is crucial for optimizing machine learning models by identifying important attributes.
• Dimensionality reduction techniques are essential for improving model efficiency and performance.
• Rough set theory provides a framework for feature reduction, particularly with discrete data.

Purpose of the Study:

• To propose and validate a novel feature selection methodology using decision reducts from rough set theory.
• To enhance attribute ranking for filtering decision rules and achieving dimensionality reduction.
• To evaluate the impact of the proposed method on classifier performance with both numerical and discrete data.

Main Methods:

• Attribute rankings were generated using a weighting factor based on decision reducts.
• The dominance relation in rough set theory was employed to handle real-valued (continuous) data.
• Discretization techniques were applied to transform numeric attributes before reduct computation.
• Decision rules were filtered based on calculated ranking scores.

Main Results:

• The proposed methodology enabled effective dimensionality reduction across various rule set configurations.
• Classifier performance, measured by predictive power, showed improvement with the applied feature selection.
• The method demonstrated successful application to continuous, discretized, and discrete attributes.

Conclusions:

• The developed feature selection technique based on decision reducts effectively reduces dimensionality.
• The methodology leads to enhanced predictive accuracy in rule-based classification tasks.
• The integration of rough set theory with dominance relations offers a robust approach for feature selection in mixed data types.