Analysis of different discriminant measures on a penalized mix-norm classification method for ERP detection

VICTORIA PETERSON; HUGO L. RUFINER; RUBÉN SPIES

Chubut

Congreso; VI MACI 2017 - Sexto Congreso de Matemática Aplicada, Computacional e Industrial; 2017

A brain-computer interface (BCI) system based on event related potentials (ERPs) consists mainly of solving a binary classification problem. Although the linear discriminant analysis (LDA) method is widely used for this type of problems, it does not yield satisfactory performances when the number of features is large relative to the number of observations. In this article we present a generalized sparse discriminant analysis method and analyze the impact of six different discriminant measures (used in the construction of certain anisotropy matrices) in classification performance. Numerical results indicate that the best measures for this type of ERP classification problems are those belonging to the Shannon-Entropy family.