A functional iterative approach for twin bounded support vector machine with squared pinball loss (Spin-FITBSVM).

Area of Science:

• Machine Learning
• Computational Intelligence
• Data Mining

Background:

• Twin Support Vector Machines (TSVM) offer lower learning costs for binary classification compared to SVM.
• However, TSVM and SVM are sensitive to noise and lack stability, motivating the development of more robust algorithms.

Purpose of the Study:

• To propose a novel functional iterative approach for twin-bound SVM with squared pinball loss (Spin-FITBSVM).
• To enhance robustness, strong convexity, and resampling stability in binary classification algorithms.
• To reduce the time complexity and eliminate the need for external optimization toolboxes.

Main Methods:

• A new functional iterative approach is developed to solve twin-bound SVM with squared pinball loss.
• The method avoids solving pairs of dual quadratic programming problems, reducing computational load.
• Numerical experiments are conducted on diverse datasets to evaluate performance.

Main Results:

• The proposed Spin-FITBSVM demonstrates improved robustness, strong convexity, and stability.
• It achieves reduced time complexity compared to traditional TSVM methods.
• Experimental results validate the superiority of Spin-FITBSVM over baseline and recent models on noisy datasets.

Conclusions:

• Spin-FITBSVM offers a more robust and stable alternative for binary classification, particularly in the presence of noisy data.
• The functional iterative approach provides an efficient method for solving the twin-bound SVM problem.
• The proposed model shows significant applicability and improved performance in machine learning tasks.