Elastic Net Nonparallel Hyperplane Support Vector Machine and Its Geometrical Rationality.

Area of Science:

• Machine Learning
• Computational Intelligence
• Pattern Recognition

Background:

• Twin Support Vector Machine (TWSVM) uses two nonparallel hyperplanes but suffers from separate quadratic programming problems, leading to inconsistent classifiers and reduced accuracy.
• The 1-norm penalty in TWSVM lacks a clear geometrical interpretation.
• Existing methods do not fully address the geometric definition or simultaneous hyperplane construction.

Purpose of the Study:

• To propose a novel Elastic Net Nonparallel Hyperplane Support Vector Machine (ENNHSVM).
• To enhance classification consistency and prediction accuracy compared to traditional TWSVM.
• To provide a better geometrical definition and improve computational efficiency.

Main Methods:

• Developed ENNHSVM incorporating elastic net penalty for slack variables.
• Constructed two nonparallel separating hyperplanes simultaneously.
• Derived the violation tolerance upper bound and designed a safe screening rule for computational speed-up.

Main Results:

• ENNHSVM demonstrates improved consistency and prediction accuracy.
• The elastic net penalty provides a more robust geometrical definition.
• The safe screening rule significantly speeds up ENNHSVM calculations.

Conclusions:

• ENNHSVM offers a theoretically sound and practically efficient alternative to TWSVM.
• The simultaneous hyperplane construction and elastic net penalty are key to its improved performance.
• ENNHSVM shows superior performance on synthetic and benchmark datasets against various SVM variants.