Transforming Clustered Non-Linearly Separable Data using Nonlinear Mapping Functions in SVM for Enhanced Classification

Abstract

Support Vector Machines is renowned for their robustness in handling classification and regression. They work by finding the best hyperplane that is able to class data into different classes very well. But when dealing with data that is clustered as well as non-linearly separable, SVM can suffer from issues in establishing crisp decision boundaries. To overcome this, using nonlinear mapping functions is useful. These operations assist in mapping information into higher-dimensional feature spaces in which the nonlinear patterns can be specified more distinctively. Our study investigates the capability of non-linear mapping functions to transform cluster, non-linearly separable information into a feature space without increasing the level of dimension complexity. SVM are reported to be capable of distinguishing information by determining optimal hyperplanes that separate distinct classes. Using nonlinear mapping functions, we establish different linear decision boundaries in the feature space, thus improving the accuracy of classifying non-linear data. The research explores the influence of altering the parameter on such a transformation and includes comparative results for, and to prove the sustainability of the method. Further, the research advances the knowledge of SVM and kernel techniques while enabling examination of the significance of different sets of features and encouraging the creation of machine learning techniques.