Adaptive Subspace Optimization Ensemble Method for High-Dimensional Imbalanced Data Classification.

Area of Science:

• Machine Learning
• Data Science
• Computer Science

Background:

• High-dimensional imbalanced data poses significant challenges for classifier performance.
• Existing methods like resampling and cost-sensitive learning struggle with noise and redundancy in high-dimensional data.

Purpose of the Study:

• To propose an adaptive subspace optimization ensemble method (ASOEM) for effective high-dimensional imbalanced data classification.
• To overcome limitations of current approaches in handling noisy and redundant high-dimensional imbalanced datasets.

Main Methods:

• Introduced a novel adaptive subspace optimization (ASO) method, incorporating adaptive subspace generation (ASG) and rotated subspace optimization (RSO).
• ASO generates robust and discriminative subspaces for constructing accurate and diverse base classifiers.
• Applied a resampling scheme on optimized subspaces to create class-balanced data for each base classifier.

Main Results:

• ASOEM demonstrated superior performance compared to mainstream imbalance learning and classifier ensemble methods.
• Experiments conducted on 24 real-world high-dimensional imbalanced datasets validated the method's effectiveness.
• The proposed method consistently outperformed existing approaches across various resampling strategies.

Conclusions:

• The adaptive subspace optimization ensemble method (ASOEM) offers a robust solution for high-dimensional imbalanced data classification.
• ASOEM effectively addresses the limitations of existing methods by optimizing subspaces and balancing class distributions.
• The findings suggest ASOEM as a promising approach for improving classifier accuracy in challenging data scenarios.