A multiple kernel classification approach based on a Quadratic Successive Geometric Segmentation methodology with a...

Area of Science:

• Machine Learning
• Pattern Recognition
• Data Mining

Background:

• Traditional classification methods struggle with large datasets.
• Need for efficient and accurate non-linear classifiers.

Purpose of the Study:

• Introduce a novel approach for classification and learning problems.
• Develop a method to handle large datasets effectively.
• Achieve accurate non-linear classification.

Main Methods:

• Applied Successive Geometric Segmentation technique.
• Utilized Oriented Bounding Hyper Boxes (OBHBs) for data encapsulation.
• Employed linear separation analysis to define OBHBs as data abstraction layers or kernels.
• Used quadratic discriminant functions to assemble non-linear separating surfaces.

Main Results:

• Successfully encapsulated large datasets using OBHBs.
• Demonstrated accurate non-linear classification.
• Achieved high-speed linear analysis for large datasets.
• Validated the methodology on UCI Machine Learning repository and a real-world Power Transformer Fault Diagnosis problem.

Conclusions:

• The Successive Geometric Segmentation approach offers an efficient and accurate method for non-linear classification.
• OBHBs serve as effective data abstraction layers.
• The technique shows potential for various applications involving large datasets.