Classifier Ensemble Based on Multiview Optimization for High-Dimensional Imbalanced Data Classification.

Area of Science:

• Machine Learning
• Data Science
• Computer Science

Background:

• High-dimensional imbalanced data pose significant challenges for classification algorithms.
• Redundant features and class imbalance hinder optimal classifier construction.
• Classifier ensembles offer improved performance over individual classifiers.

Purpose of the Study:

• To develop a novel multiview optimization (MVO) approach for effective feature learning.
• To design an accurate and robust classifier ensemble system for high-dimensional imbalanced data.
• To enhance classification performance by addressing feature redundancy and class imbalance.

Main Methods:

• Proposed a multiview optimization (MVO) framework.
• Introduced optimized subview generation (OSG) to create diverse, effective feature subsets.
• Developed a selective ensemble of optimized subviews (SEOS) using a novel evaluation criterion.
• Applied an oversampling technique for class rebalancing on optimized views.

Main Results:

• The MVO method effectively learns robust and discriminative features.
• OSG enhances feature classification ability and ensemble diversity.
• SEOS successfully selects optimal subviews for ensemble construction.
• Experimental results demonstrate superior performance compared to mainstream ensemble methods on 25 datasets.

Conclusions:

• The proposed MVO approach offers a powerful solution for high-dimensional imbalanced classification.
• The method effectively tackles feature redundancy and class imbalance issues.
• This work advances classifier ensemble techniques for complex datasets.