PHFS: Progressive Hierarchical Feature Selection Based on Adaptive Sample Weighting.

Area of Science:

• Machine Learning
• Data Science
• Artificial Intelligence

Background:

• Hierarchical feature selection is vital for high-dimensional data with complex label structures.
• Existing methods struggle with label noise and lack adaptive sample weighting, limiting their effectiveness.
• Mitigating label noise is crucial for accurate feature selection in hierarchical datasets.

Purpose of the Study:

• To propose an adaptive sample weighting-based progressive hierarchical feature selection (PHFS) method.
• To enhance the effectiveness of feature selection in the presence of label noise.
• To improve performance on complex hierarchical data with numerous classes.

Main Methods:

• PHFS integrates progressive sample selection and hierarchical feature selection.
• It dynamically adjusts sample weights to prioritize high-quality data.
• Employs a two-stage progressive selection process with adaptive weighting and matrix factorization.

Main Results:

• PHFS effectively reduces the impact of label noise by focusing on correctly labeled samples.
• The method demonstrated superior performance compared to 13 state-of-the-art techniques.
• Extensive experiments on eight real-world datasets validated PHFS's effectiveness.

Conclusions:

• PHFS offers a robust solution for feature selection in noisy hierarchical data.
• The adaptive weighting and progressive selection enhance data quality and model performance.
• PHFS represents a significant advancement in handling label noise for hierarchical classification.