Progressive Semisupervised Learning of Multiple Classifiers.

Area of Science:

• Machine Learning
• Artificial Intelligence
• Data Science

Background:

• Semisupervised learning is crucial for datasets with few labeled samples.
• Conventional methods struggle with high-dimensional data and lack training set expansion.
• Existing approaches often fail to optimize training set enlargement.

Purpose of the Study:

• To introduce a progressive semisupervised ensemble learning approach (PSEMISEL).
• To address limitations of conventional semisupervised ensemble learning on high-dimensional, low-label datasets.
• To enhance training set size through optimization processes.

Main Methods:

• Utilizes the random subspace technique to explore dataset structures in subspaces.
• Employs a progressive training set generation process.
• Incorporates a self-evolutionary sample selection process for training set enlargement.

Main Results:

• PSEMISEL demonstrates effectiveness on most real-world datasets.
• The proposed method outperforms state-of-the-art approaches on 10 out of 18 datasets.
• Nonparametric tests confirm the superiority of PSEMISEL over other semisupervised ensemble methods.

Conclusions:

• PSEMISEL effectively handles datasets with very small numbers of labeled samples.
• The random subspace technique and progressive training set enlargement are key to PSEMISEL's success.
• This approach offers a significant advancement for semisupervised learning in challenging data scenarios.