Optimization of Nonlinear Kernel-Based Classification using Pin-Sgtbsvm

Abstract

When it comes to machine learning, the selection of kernel functions is crucial for classification model performance. Because they can simulate complicated patterns in data, nonlinear kernel-based classification algorithms have attracted a lot of interest for optimization. Using the German, Haberman, CMC, Fertility, WPBC, Ionosphere, and Live Disorders benchmark datasets as well as others from the UCI database, the paper assesses how well the Pin-SGTBSVM algorithm performs. Using a tenfold cross-validation technique, the ideal parameters are obtained using a nonlinear kernel function. Over six datasets, the findings show that Pin-SGTBSVM outperforms well-known algorithms like TWSVM, TBSVM, Pin-GTWSVM, and Pin-GTBSVM in terms of accuracy, with noticeable advances in classification performance. Although it also shows competitive results, Pin-SGTWSVM's accuracy on the German dataset is marginally worse than TWSVM's. The experimental results show that Pin-SGTBSVM is a reliable method for improving classification accuracy with no impact on computing efficiency. The results show that it might be used for data categorization and machine learning in the actual world.