Weighted twin support vector machines with local information and its application.

Area of Science:

• Machine Learning
• Computational Intelligence
• Pattern Recognition

Background:

• Twin Support Vector Machines (TWSVM) improve upon Generalized Eigenvalue Proximal Support Vector Machines (GEPSVM) but face high computational costs.
• Existing TWSVM methods do not fully exploit similarity information between data points with identical labels, potentially limiting classification performance.

Purpose of the Study:

• To introduce Weighted Twin Support Vector Machines with Local Information (WLTSVM), a novel nonparallel plane classifier.
• To address the computational cost and limited data similarity exploitation issues of previous TWSVM variants.

Main Methods:

• WLTSVM is proposed as a new classification method that integrates local information to mine sample similarities.
• The method retains the concept of support vectors, similar to standard SVM.
• It involves solving a single optimization problem, reducing computational complexity compared to TWSVM.

Main Results:

• WLTSVM demonstrates comparable or superior classification accuracy to SVM, GEPSVM, and TWSVM.
• The method is computationally more efficient than TWSVM.
• Experiments on simulated and real-world datasets validate the effectiveness of WLTSVM.

Conclusions:

• WLTSVM effectively leverages local information to enhance classification performance.
• The proposed method offers a more efficient and accurate alternative to existing twin support vector machine approaches.
• WLTSVM successfully balances classification accuracy with reduced computational demands.