Novel Ensemble Approach with Incremental Information Level and Improved Evidence Theory for Attribute Reduction.

Area of Science:

• Machine Learning
• Data Science
• Artificial Intelligence

Background:

• Intelligent technology drives exponential data growth, making attribute extraction from complex datasets challenging.
• Existing attribute reduction methods often overlook attribute redundancy, focusing solely on correlation with labels.

Purpose of the Study:

• To propose an ensemble approach for attribute reduction that addresses both attribute relevance and redundancy.
• To develop a novel method, incremental information level and improved evidence theory (IILE), for more effective attribute subset selection.

Main Methods:

• Utilizing an incremental information level reduction measure to assess attributes based on reduction capability and redundancy.
• Employing improved evidence theory and approximate reduction methods to fuse multiple reduction results.
• Achieving an approximately globally optimal and representative subset of attributes.

Main Results:

• Experimental comparisons on eight datasets demonstrate the superiority of the proposed IILE approach over existing methods.
• The IILE method successfully obtains more relevant attribute sets by effectively managing redundancy.

Conclusions:

• The proposed IILE approach offers a significant advancement in attribute reduction for complex datasets.
• This method enhances data analysis by providing a more representative and optimized subset of attributes.