On the benefits of classical multidimensional scaling in Epileptic seizure prediction studies

Abstract

Algorithms for Epileptic seizure prediction using various features extracted from the multichannel Electroencephalo-graphic (EEG) signals, need to work in high dimensional spaces, leading to increased difficulties in computational time and convergence conditions. Multidimensional Scaling (MDS) is a technique to surpass this curse of dimensionality in classification problems. In this work we investigate the influence of dimensional reduction in classification performance by previously applying Multidimensional Scaling and then applying Support Vector Machines (SVM) to classify the brain state. Data from five patients of the European Database on Epilepsy of the FP7 EPILEPSIAE Project is used. The results show that dimension reduction improves less than expected the SVM performance.