Class proximity measures--dissimilarity-based classification and display of high-dimensional data.

Area of Science:

• Machine Learning
• Data Visualization
• Bioinformatics

Background:

• High-dimensional data presents significant challenges for classification and visualization.
• Existing methods may not offer a unified approach for simultaneous analysis.
• The need for effective tools to interpret complex datasets is critical.

Purpose of the Study:

• To introduce and construct Class Proximity Planes for two-class problems.
• To extend previous relative distance plane mapping for a more general approach.
• To enable simultaneous classification and visualization of many-feature datasets.

Main Methods:

• Developed mappings of high-dimensional instances into dissimilarity (distance)-based Class-Proximity Planes.
• Utilized two-dimensional coordinate systems where axes represent distances to class proximity measures.
• Applied various Class Proximity Projections and their combinations.

Main Results:

• Demonstrated the ability to classify and visualize high-dimensional instances effectively.
• Compared classification and visualization outcomes across multiple datasets.
• Showcased the utility on UCI datasets and a high-dimensional biomedical dataset.

Conclusions:

• Class Proximity mappings offer a unified and generalizable framework.
• These projections provide diverse perspectives for dataset analysis.
• The method proves effective for both general and specialized high-dimensional data.