Robust and Sparse Principal Component Analysis With Adaptive Loss Minimization for Feature Selection.

Area of Science:

• Machine Learning
• Data Science
• Computer Vision

Background:

• Principal Component Analysis (PCA) is a widely used unsupervised subspace learning technique.
• Conventional PCA struggles with data outliers and cannot perform feature selection.
• Existing robust PCA models address outliers but lack feature selection capabilities.

Purpose of the Study:

• To propose a novel robust PCA (RPCA) model for simultaneous outlier mitigation and feature selection.
• To develop an effective method for dimensionality reduction that also identifies important features.

Main Methods:

• The proposed RPCA model utilizes the sigma-norm for reconstruction error, enhancing robustness.
• A l2,0-norm constraint is applied to subspace projection for feature selection.
• An efficient iterative optimization algorithm is developed to solve the objective function with non-convex and non-smooth constraints.

Main Results:

• The novel RPCA model effectively mitigates the impact of outliers in datasets.
• The model successfully performs feature selection alongside dimensionality reduction.
• Extensive experiments on real-world datasets validate the model's effectiveness and superiority.

Conclusions:

• The proposed RPCA model offers a robust solution for dimensionality reduction with integrated feature selection.
• This approach addresses limitations of conventional PCA in handling outliers and performing feature selection.
• The method demonstrates significant improvements in performance on various real-world datasets.