TSVR: an efficient Twin Support Vector Machine for regression.

Area of Science:

• Machine Learning
• Computational Statistics

Background:

• Classical Support Vector Regression (SVR) exhibits slow learning speeds due to complex constraint optimization.
• SVR's computational demands stem from minimizing a convex quadratic function with numerous linear inequality constraints.

Purpose of the Study:

• To introduce Twin Support Vector Regression (TSVR), a novel regression technique.
• To enhance the learning speed and generalization capabilities of Support Vector Regression.

Main Methods:

• TSVR formulates two smaller, related Support Vector Machine (SVM)-type problems.
• It determines a pair of epsilon-insensitive upper and lower bound functions.
• The approach is inspired by Twin Support Vector Machines (TSVM) using two nonparallel planes.

Main Results:

• TSVR demonstrates significantly faster learning speeds compared to classical SVR.
• Experimental results show that TSVR achieves good generalization performance.
• The method proved effective on both artificial and benchmark datasets.

Conclusions:

• TSVR presents an efficient and effective alternative to traditional SVR.
• The proposed method balances computational speed with strong predictive accuracy.
• TSVR offers a promising direction for regression tasks in machine learning.