DATA FUSION FOR PHYSIOLOGICAL VITAL SIGNS WITH AUTOMATIC EXTREME VALUE CONTROL

Abstract

In this paper, we propose methods for fusing multivariate physiological vital sign data observed over a common time into univariate data. The approaches rely on noncentral moment-based statistics derived from the physiological vital sign random variables. Thus preserve the vital sign specific auto-correlations and appropriately control the within-vital sign extreme observations automatically so that such observations are utilized in the model specification without being deleted. Data mixing factors and weights are developed and allowed to depend on the fusion statistics so as to inherit their appealing features. This way, a single model can be fitted to the composite data instead of multiple models for the unfused data resulting in computational and memory savings. Implementation of the approaches using real vital sign data illustrates the utility of the schemes in extracting appropriate univariate data from the multivariate data that can be handled in terms of the common time within the functional regression framework.