Tensor-based Multi-view Feature Selection with Applications to Brain Diseases.

Area of Science:

• Multi-view learning
• Machine learning in medical science
• Big data analytics

Background:

• Multi-view learning leverages complementary information from diverse data sources for enhanced learning tasks.
• In medical science, integrating data from various examinations (clinical, imaging, etc.) is crucial for comprehensive subject analysis.
• Combining features from multiple views can introduce noise and irrelevant information, necessitating feature selection.

Purpose of the Study:

• To develop an effective feature selection method for multi-view learning in medical diagnosis.
• To address the challenge of irrelevant information introduced by combining multiple data views.
• To improve the accuracy and relevance of selected features for disease diagnosis.

Main Methods:

• Exploration of tensor product to integrate different data views into a joint space.
• Development of a dual method of tensor-based multi-view feature selection (dual-Tmfs).
• Application of support vector machine recursive feature elimination principles within the dual-Tmfs framework.

Main Results:

• The proposed dual-Tmfs method demonstrated superior classification performance on neurological disorder datasets.
• Features selected by dual-Tmfs were found to be highly relevant for disease diagnosis.
• Effective integration of multi-view data leading to improved diagnostic accuracy.

Conclusions:

• Tensor-based multi-view feature selection is a promising approach for medical big data analysis.
• The dual-Tmfs method offers an effective solution for selecting relevant features in multi-view learning.
• This approach enhances disease diagnosis by improving classification performance and feature relevance.