Abstracting steady qualitative descriptions over time from noisy, high-frequency data

Abstract

On-line monitoring at neonatal intensive care units produces high volumes of data. Numerous devices generate data at high frequency (one data set every second). Both, the high volume and the quite high error-rate of the data make it essential to reach at higher levels of description from such raw data. These abstractions should improve the medical decision making. We will present a time-oriented data-abstraction method to derive steady qualitative descriptions from oscillating high- frequency data. The method contains tunable parameters to guide the sensibility of the abstraction process. The benefits and limitations of the different parameter settings will be discussed.