A non-Gaussian approach for biosignal classification based on the Johnson SU translation system

This paper proposes a non-Gaussian approach for biosignal classification based on the Johnson S_U translation system. The Johnson system is a normalizing translation that transforms data without normality to normal distribution using four parameters, thereby enabling the representation of a wide range of shapes for marginal distribution with skewness and kurtosis. In this study, a discriminative model based on the multivariate Johnson S_U translation system is transformed into linear combinations of coefficients and input vectors using log-linearization, and is incorporated into a neural network structure, thereby allowing the determination of model parameters as weight coefficients of the network via backpropagation-based training. In the experiments, the classification performance of the proposed network is demonstrated using artificial data and electromyogram data.