A novel support vector machine with globality-locality preserving.

Area of Science:

• Machine Learning
• Pattern Recognition
• Data Mining

Background:

• Standard Support Vector Machines (SVM) are effective for classification but can be sensitive to class distribution, leading to nonrobust solutions.
• The primal optimal model of SVM may not preserve the intrinsic manifold structure of the data space.
• Addressing these limitations is crucial for developing more reliable classification algorithms.

Purpose of the Study:

• To propose an improved Support Vector Machine (SVM) model named GLPSVM.
• To enhance SVM's robustness by incorporating globality-locality preserving principles.
• To evaluate the effectiveness of GLPSVM in preserving data manifold structure during classification.

Main Methods:

• Introduction of globality-locality preserving (GLP) into the standard SVM framework.
• Development of the Globality-Locality Preserving Support Vector Machine (GLPSVM) model.
• Experimental validation using diverse datasets from the UCI machine learning repository.

Main Results:

• GLPSVM demonstrated superior performance compared to standard SVM and its variants.
• Exceptional recognition rates exceeding 97% were achieved on the Wine and Iris datasets.
• The model effectively preserved the manifold structure of the data space, contributing to improved classification accuracy.

Conclusions:

• GLPSVM offers a robust and effective approach to pattern classification.
• The integration of globality-locality preserving principles significantly enhances SVM performance.
• GLPSVM represents a notable advancement over existing SVM-based algorithms, particularly for datasets with complex structures.