Unveiling social network clans: improving genealogical clan classification with SVM neural classifiers and enhanced kernels

Abstract

In this study, we developed a variant of the support vector machine (SVM) neural classifier and utilized it to categorize clans in a genealogical dataset. For each of the five kernels, all four variants, twin SVM (TSVM), proximal SVM (PSVM), twin proximal SVM (TPSVM), and multi-class SVM (MCSVM) classifier are simulated and tested. The analysis of variance - radial basis function (ANOVA RBF) kernel outperformed all other SVM variants, in terms of classification accuracy with the lowest error value. Additionally, it is found that for the considered dataset, TPSVM neural classifier with ANOVA RBF Kernel generated 98.91% classification accuracy, and the TPSVM classifier has achieved the minimized mean square error (MSE) value of 0.00015. The Twin Proximal SVM classifier has produced enhanced classification accuracy with better precision and F1-score in comparison to all other developed and simulated SVM classifier models.