Bayesian classifier with multivariate distribution based on D-vine copula model for awake/drowsiness interpretation during power nap

Abstract

In this study, a Bayesian classifier with the multivariate distribution based on the D-vine copula model is developed and evaluated for the awake/drowsiness interpretation during the power nap. The objective is to consider the correlation among the features into the automatic classification algorithm. A power nap is a short sleep process, which is commonly considered as a supplement to the insufficient overnight sleep. It may involve the states of awake and drowsiness. Neurophysiological features are extracted from the EEGs (electroencephalography) and EOGs (electrooculography), which are synchronously recorded during one's short nap after lunch. The multivariate distribution of features is decomposed into independency and dependency products according to the D-vine copula model. The independency product is the marginal probability density function of the features. The dependency product consists of pair-copula functions. The marginal probability density is estimated by the kernel function and k-nearest-neighbor density respectively. The parameters of pair-copula functions are estimated by the maximum likelihood estimation. In total, 8 healthy subjects were involved. The comparison results showed that the Bayesian classifier with the multivariate distribution based on the D-vine copula model obtained quite satisfied classification accuracy. The developed method introduced a feasible way to construct the multivariate distribution, which can enhance the classification performance of Bayesian classifier when dealing with the complex correlation of features in actual cases.