Handling Multi-Class Problem by Intuitionistic Fuzzy Twin Support Vector Machines Based on Relative Density...

Area of Science:

• Machine Learning
• Data Mining
• Pattern Recognition

Background:

• Intuitionistic fuzzy twin support vector machine (IFTSVM) combines intuitionistic fuzzy sets (IFS) and twin support vector machines (TSVM) to mitigate noise and outliers.
• Existing IFTSVM methods struggle with multi-class problems, high-dimensional data, and exhibit high computational complexity due to score function calculations.

Purpose of the Study:

• To propose a novel IFTSVM variant that addresses limitations of existing methods.
• To enhance classification accuracy and computational efficiency for multi-class and high-dimensional datasets.

Main Methods:

• Introduced relative density estimation by approximating probability distributions using K-nearest-neighbor distances.
• Employed a one-versus-one-versus-rest strategy for k-class hyperplane construction.
• Utilized a coordinate descent system to reduce computational complexity during training.

Main Results:

• The proposed method demonstrated promising performance across 24 benchmark datasets.
• Statistical validation using bootstrap resampling (95% confidence interval) and Friedman tests confirmed significant performance improvements.
• The enhanced IFTSVM showed superiority over other support vector machine models in numerical evaluations.

Conclusions:

• The novel IFTSVM approach effectively handles multi-class and high-dimensional classification problems.
• Relative density estimation and coordinate descent significantly improve computational efficiency and accuracy.
• This work offers a robust and efficient alternative for complex classification tasks in machine learning.