Sparse PCA via l_2,p-Norm Regularization for Unsupervised Feature Selection.

Area of Science:

• Data Mining
• Machine Learning
• Dimensionality Reduction

Background:

• High-dimensional data presents challenges in data mining.
• Unsupervised feature selection is crucial for analyzing unlabeled data.
• Existing spectral methods are sensitive to noise and computationally expensive.

Purpose of the Study:

• To propose a simple and efficient unsupervised model for feature selection.
• To address the limitations of spectral-based methods in handling noisy and large datasets.

Main Methods:

• Formulating Principal Component Analysis (PCA) as a reconstruction error minimization problem.
• Incorporating l2,p-norm regularization to achieve a sparse projection matrix.
• Developing an efficient optimization algorithm with theoretical analysis of convergence and complexity.

Main Results:

• The proposed method learns a row-sparse and orthogonal projection matrix.
• Experimental results on synthetic and real-world datasets demonstrate effectiveness.
• The method successfully selects discriminative features from noisy data.

Conclusions:

• The developed unsupervised model offers an effective solution for feature selection in high-dimensional data.
• The method is efficient and robust to noise, outperforming existing techniques.
• This work contributes to advancing unsupervised learning for data analysis.