Diversity-based selection of components for fusion classifiers.

Area of Science:

• Computer Science
• Machine Learning
• Pattern Recognition

Background:

• Fusion classifiers, vital for clinical diagnostics and pattern recognition, rely heavily on component diversity for optimal performance.
• Assessing and selecting diverse components is a critical challenge in designing effective fusion systems.

Purpose of the Study:

• Introduce a novel pairwise diversity-based ranking strategy for selecting diverse ensemble components.
• Develop unified classifier and data fusion systems based on this diversity selection strategy.

Main Methods:

• A new pairwise diversity-based ranking strategy was developed to identify subsets of ensemble components with maximal diversity.
• Classifier fusion and data fusion systems were formulated using the proposed diversity selection strategy.
• The strategy was applied to classify multi-channel event-related potentials (ERPs).

Main Results:

• Data fusion systems demonstrated superior performance compared to classifier fusion systems.
• The diversity-based data fusion system significantly outperformed systems using randomly selected data components.
• The diversity selection strategy effectively identified optimal data component combinations for enhanced classification performance.

Conclusions:

• Diversity-based component selection is essential for improving the performance of fusion classifiers.
• Data fusion, particularly when components are selected using a diversity metric, offers a more effective approach than classifier fusion.
• The proposed strategy provides a robust method for optimizing component selection in data fusion systems for pattern recognition tasks.