Semi-supervised linear discriminant clustering.

Area of Science:

• Machine Learning
• Data Mining
• Pattern Recognition

Background:

• Semi-supervised learning leverages both labeled and unlabeled data.
• Clustering and dimensionality reduction are key tasks in data analysis.
• Existing methods often struggle with limited labeled data.

Purpose of the Study:

• To develop a semi-supervised clustering algorithm that integrates dimensionality reduction.
• To enhance clustering performance by effectively utilizing unlabeled data through soft labels.
• To create a flexible framework adaptable to different soft label estimation techniques.

Main Methods:

• Proposed semi-supervised linear discriminant clustering (Semi-LDC) algorithm.
• Simultaneous K-means clustering and linear discriminant analysis (LDA).
• Utilized constrained-PLSA for estimating soft labels of unlabeled data.
• Employed soft LDA with both hard and soft labels to find a projection matrix.

Main Results:

• Semi-LDC demonstrated superior performance compared to other semi-supervised methods across three datasets.
• The use of soft labels significantly improved clustering performance, particularly when labeled data was scarce.
• Analysis confirmed the positive influence of soft labels on classification accuracy.

Conclusions:

• Semi-LDC offers an effective approach for semi-supervised clustering with integrated dimensionality reduction.
• Soft labels are crucial for improving model robustness and accuracy in low-label scenarios.
• The proposed method serves as a versatile framework adaptable to various soft label estimation strategies.