Intuitionistic proximity-consensus fuzzy geometric twin support vector machine with best subset selection

Area of Science:

• Machine Learning
• Pattern Recognition
• Data Mining

Background:

• Intuitionistic fuzzy twin support vector machine (IFTSVM) enhances robustness but struggles with resampling noise and feature sensitivity.
• Existing methods like IFTSVM have limitations in handling specific noise types and feature redundancy.

Purpose of the Study:

• To propose a novel intuitionistic proximity-consensus fuzzy (IPCF) geometric TSVM.
• To simultaneously reduce outliers, resampling noise, and noisy features using l0-norm best subset selection.
• To develop an efficient optimization algorithm for the proposed model.

Main Methods:

• The proposed IPCF is induced by distance to the hyperplane from least squares one-class SVM, unlike IFS's distance to class center.
• Incorporation of l0-norm penalty for simultaneous feature selection and noise feature reduction.
• Development of an intuitionistic proximity-consensus fuzzy-optimized variable sorted active set algorithm (IPCF-VSAS) to solve the non-convex optimization problem.

Main Results:

• Experimental results on simulated and UCI datasets validate the proposed method's advantages.
• The IPCF-TSVM method demonstrates superior performance on datasets with redundant features and noise compared to state-of-the-art support vector classifiers.
• The l0-norm penalty effectively performs feature selection and noise feature reduction.

Conclusions:

• The proposed IPCF-TSVM with l0-norm best subset selection effectively addresses limitations of previous methods.
• The IPCF-VSAS algorithm provides an efficient solution for optimizing the proposed model.
• The method shows significant potential for improving classification accuracy in noisy and feature-rich datasets.