Sparse Trace Ratio LDA for Supervised Feature Selection.

Area of Science:

• Data Mining
• Machine Learning
• Pattern Recognition

Background:

• High-dimensional data presents challenges for classification tasks.
• Dimensionality reduction, including feature extraction and selection, is crucial for improving classification performance.
• Linear Discriminant Analysis (LDA) is a classic feature extraction method known for category discrimination.

Purpose of the Study:

• To propose a novel supervised feature selection method that integrates the discriminative power of LDA with the advantages of feature selection.
• To address the performance degradation of classification in high-dimensional datasets.
• To develop an effective approach for selecting discriminative features.

Main Methods:

• A supervised feature selection method is proposed, combining trace ratio LDA with L2,p-norm regularization.
• An orthogonal constraint is imposed on the projection matrix.
• An optimization algorithm is developed to solve the proposed method.

Main Results:

• The proposed method learns a row-sparse projection matrix for selecting discriminative features.
• Extensive experiments on synthetic and real-world datasets demonstrate the method's effectiveness.
• The integrated approach shows superior performance in high-dimensional data classification.

Conclusions:

• The proposed supervised feature selection method effectively handles high-dimensional data for classification.
• Integrating LDA's discrimination with feature selection advantages offers a powerful approach.
• The method shows significant potential for improving classification accuracy in various applications.