Improving Classification in Support Vector Machine Using Neutrosophic Logic

Support Vector Machine (SVM) is considered one of the most effective methods for classification tasks. However, randomness in classifying observations on the optimal separating hyperplane decision — which are often indeterminate in their class membership — and misclassified observations remain among the main challenges that researchers face when using SVM. To address these challenges, a Neutrosophic logic – based approach was proposed, enabling better handling of class ambiguity and reducing misclassifications in the neighborhood of the optimal separating hyperplane. A novel algorithm was introduced, which consisted of two main steps. First, the observations located near the optimal separating hyperplane were converted into Neutrosophic data, characterized by three components: truth, indeterminacy, and falsity. In the second step, the SVM classifier was reapplied to the Neutrosophic data to improve the classification accuracy. As a case study, the proposed algorithm was tested on two types of oil samples (sunflower oil and corn oil), and its performance was compared with that of a standard SVM without Neutrosophic data. The results demonstrated that the SVM models utilizing Neutrosophic data achieved higher accuracy compared to those without Neutrosophic data. Therefore, integrating Neutrosophic logic with SVM can significantly enhance the performance and reliability of SVM based models.