Inverse free reduced universum twin support vector machine for imbalanced data classification.

Area of Science:

• Machine Learning
• Data Science
• Artificial Intelligence

Background:

• Imbalanced datasets are common in real-world scenarios, posing challenges for standard classification algorithms.
• Traditional algorithms like Twin Support Vector Machines (TSVM) perform poorly on imbalanced data.
• Existing improvements such as Universum Twin Support Vector Machine (UTSVM) and Reduced Universum Twin Support Vector Machine (RUTSVM) still face computational drawbacks.

Purpose of the Study:

• To address the limitations of existing methods for imbalanced dataset classification.
• To propose an improved algorithm, Improved Reduced Universum Twin Support Vector Machine (IRUTSVM), that overcomes computational inefficiencies.
• To enhance the generalization performance of algorithms on imbalanced datasets.

Main Methods:

• Developed an Improved Reduced Universum Twin Support Vector Machine (IRUTSVM) by modifying Lagrangian functions.
• Integrated a term from the objective function into constraints to avoid matrix inversion in primal problems.
• Utilized smaller rectangular kernel matrices to reduce computational time.

Main Results:

• The proposed IRUTSVM eliminates the need for matrix inversion during training and classifier computation.
• IRUTSVM demonstrates superior generalization performance compared to TSVM, UTSVM, and RUTSVM.
• Extensive testing on synthetic and real-world imbalanced datasets validates the algorithm's effectiveness.

Conclusions:

• IRUTSVM offers a computationally efficient and effective solution for imbalanced dataset classification.
• The modified approach significantly improves upon existing TSVM-based methods for imbalanced learning.
• IRUTSVM presents a promising advancement in machine learning for real-world imbalanced data problems.