Unsupervised Feature Selection With Weighted and Projected Adaptive Neighbors.

Area of Science:

• Data Mining
• Machine Learning
• Artificial Intelligence

Background:

• High-dimensional data presents computational challenges and hinders satisfactory results in data mining.
• Unsupervised feature selection aims to reduce data dimensionality without labels by preserving intrinsic data structure.
• Existing methods may fail due to noisy features affecting the reliability of the intrinsic data structure.

Purpose of the Study:

• To propose a new unsupervised feature selection model robust to noisy features.
• To simultaneously learn graph structure, feature weights, and projection matrix for reliable intrinsic structure construction.
• To enhance the performance of feature selection in high-dimensional datasets.

Main Methods:

• A novel unsupervised feature selection model is developed.
• Simultaneous learning of graph structure, feature weights, and projection matrix.
• Incorporation of adaptive neighbors and a constraint for exactly c connected components in the graph.
• An optimization algorithm is presented to solve the model, with analysis of parameter determination and computational complexity.

Main Results:

• The proposed method effectively addresses the issue of noisy features in unsupervised feature selection.
• Experimental results on synthetic and real-world datasets demonstrate the method's effectiveness.
• The simultaneous learning approach leads to a more reliable intrinsic data structure.

Conclusions:

• The novel unsupervised feature selection model offers a robust solution for high-dimensional data.
• The method enhances data mining by improving feature selection in the presence of noise.
• The approach of simultaneously learning graph structure, feature weights, and projection is validated by experimental evidence.