High Order Moment Features for NIRS-Based Classification Problems

Abstract

This paper aims to experiment high order moment features in two well-known problems which are motor imagery and person authentication in Brain Computer Interface (BCI) systems using Near Infrared Spectroscopy (NIRS) technique. To improve performance of the systems, we propose a new feature by combining 2^nd order and 4^th order moments of signal together. Our results show that such the feature not only achieves very high recall and precision ratios but also is practical for online NIRS-based BCI systems. Our systems can achieve recall and precision ratio at 99.2% for the left-hand and right-hand imagery problem, and up to 100% for the person authentication problem.