Improving SVM performance through data reduction and misclassification analysis with linear programming

Area of Science:

• Machine Learning
• Optimization
• Computational Science

Background:

• Support Vector Machines (SVM) involve complex dual optimization problems where each data point represents a decision variable.
• High dimensionality in SVM optimization can lead to computational inefficiencies.
• Understanding data separability and misclassification rates is crucial for effective classification.

Purpose of the Study:

• To develop efficient linear programming models for Support Vector Machine (SVM) optimization.
• To introduce methods for determining linear separability and computing misclassification rates.
• To reduce the dimensionality of SVM optimization problems through data reduction techniques.

Main Methods:

• Formulating linear programming models to assess linear separability and calculate misclassification rates.
• Utilizing a convexity property for data reduction in linearly separable cases.
• Integrating SVM optimization with linear programming for a combined analysis framework.

Main Results:

• Demonstrated efficient methods for determining linear separability of data sets.
• Established the misclassification rate as a key metric for classification complexity.
• Showcased data reduction techniques to improve SVM optimization efficiency.

Conclusions:

• Linear programming offers an efficient approach to optimize Support Vector Machines by reducing dimensionality.
• The proposed methods provide a comprehensive framework for classification and complexity analysis.
• Data reduction and misclassification rate analysis enhance the practical application of SVMs.