Multi-parametric trends analysis and events prediction in the context of a cardiac rehabilitation system

Abstract

The final goal of this work is the development of methodologies for time series prediction. The aim is to support the forecast of biosignals collected by telemonitoring systems, as well as the early detection of critical events (such as hypertension episodes, based on the evolution of blood pressure). The main hypothesis is that the estimation of biosignals future evolution can be supported on current and past measurements taken from historical data. To this end, two main stages are considered: similarity analysis, to find a set of similar patterns in the historical dataset followed by a prediction mechanism, based on the obtained patterns. Two approaches are proposed based on this principle: singleparametric and multi-parametric ones. The validation and comparison of both single and multiparametric approaches is performed using blood pressure, heart rate and body weight signals collected during the myHeart telemonitoring study.