Small sphere and large margin support tensor machines for imbalanced tensor data classification.

Area of Science:

• Machine Learning
• Data Science
• Artificial Intelligence

Background:

• Imbalanced data classification is challenging, especially with tensor data.
• Existing methods like the small sphere and large margin approach (SSLM) are limited to vector data.

Purpose of the Study:

• To propose a novel model for imbalanced tensor data classification.
• To leverage the structural information inherent in tensor data for improved classification.

Main Methods:

• Introduction of the small sphere and large margin support tensor machine (SSLMSTM).
• Construction of two concentric hyperspheres centered by a rank-1 tensor.
• Extension to higher rank R cases (HR-SSLMSTM).
• Utilizing CANDECOMP/PARAFAC decomposition and alternating iteration for model solving.

Main Results:

• SSLMSTM effectively captures normal samples within a small hypersphere and pushes outliers outside a large hypersphere.
• Increased margin between hyperspheres enhances performance.
• Experiments demonstrate the validity and effectiveness of both SSLMSTM and HR-SSLMSTM.

Conclusions:

• SSLMSTM and HR-SSLMSTM offer a robust solution for imbalanced tensor data classification.
• The proposed models successfully utilize tensor data structure for superior performance.
• This work opens new avenues for handling complex, imbalanced datasets in machine learning.