Adaptive Semi-Supervised Classifier Ensemble for High Dimensional Data Classification.

Area of Science:

• Data Mining
• Machine Learning
• Artificial Intelligence

Background:

• High dimensional data classification with scarce labeled data presents significant challenges.
• Existing semi-supervised classifier ensemble (SSCE) methods often struggle with such data limitations.

Purpose of the Study:

• To propose and evaluate an adaptive semi-supervised classifier ensemble framework (ASCE) for high dimensional data classification.
• To enhance classification performance by incorporating adaptive processes for feature selection, sample weighting, and auxiliary training data generation.

Main Methods:

• Developed a feature selection-based semi-supervised classifier ensemble framework (FSCE).
• Introduced an adaptive semi-supervised classifier ensemble framework (ASCE) building upon FSCE.
• ASCE incorporates adaptive feature selection, adaptive weighting process (AWP), and auxiliary training set generation process (ATSGP).
• Employed nonparametric tests for comparing SSCE approaches across diverse datasets.

Main Results:

• Experiments on 20 high dimensional real-world datasets demonstrated the efficacy of ASCE.
• The adaptive processes within ASCE significantly improved the performance of the SSCE approach.
• ASCE showed robust performance on datasets with very limited labeled training data.

Conclusions:

• ASCE effectively addresses the challenge of high dimensional data classification with limited labeled data.
• The adaptive components of ASCE are crucial for enhancing classification performance.
• ASCE outperforms most state-of-the-art SSCE approaches in high dimensional settings.