Non-Linear Sequential SVM Classifier of Epileptic Seizures

Abstract

This paper concerns a design for implementing Support Vector Machine (SVM), with non-linear Gaussian kernel on Field Programmable Gate Array (FPGA), for development of an accurate seizure epilepsy classification. The proposed methodology depends on storing the extracted support vectors, along with the SVM parameters into Lookup Tables. The proposed SVM architecture depends on feeding the selected support vectors into a single Gaussian kernel core in a sequential fashion, rather than feeding them parallely to the kernel cores, thus reducing the resources usage on the target FPGA board. The system is implemented on Xilinx Virtex6 xc6vcx75t board. System verifications and simulations have been done. The proposed methodology achieves accuracy of 88.53%, along with average sensitivity and specificity of 86.4% and 90.83% respectively.