Kernel Partial Least Squares Feature Selection Based on Maximum Weight Minimum Redundancy.

Area of Science:

• Machine Learning
• Data Mining
• Pattern Recognition

Background:

• Feature selection is crucial for machine learning and data mining, aiming to identify important features while minimizing redundancy.
• Existing methods like maximum weight minimum redundancy (mRMR) have limitations with diverse datasets and high-dimensional data challenges.
• Adapting feature evaluation criteria is necessary for optimal performance across various data characteristics.

Purpose of the Study:

• To propose a novel kernel partial least squares (KPLS) feature selection method.
• To enhance the maximum weight minimum redundancy (mRMR) algorithm for improved high-dimensional data analysis.
• To improve classification accuracy and simplify calculations for complex datasets.

Main Methods:

• Developed a KPLS-based feature selection approach integrating an enhanced mRMR algorithm.
• Introduced a weight factor to adjust the balance between feature weight and redundancy in the evaluation criteria.
• The method considers feature-dataset and feature-class label redundancy for diverse data types.

Main Results:

• The proposed KPLS method demonstrated effective feature subset selection for high-dimensional datasets.
• Experimental results showed improved classification accuracy compared to other feature selection methods.
• The method proved feasible and effective across various datasets, including those with noise.

Conclusions:

• The enhanced KPLS feature selection method offers a robust solution for high-dimensional data analysis.
• It effectively addresses the limitations of traditional mRMR by incorporating dataset-specific criteria.
• The approach achieves superior classification performance, highlighting its practical utility.