Towards symbolization of intensive monitoring data for knowledge based inference systems

Computer programs have been built for automatic collection of all intensively monitored physiological data, their storage and retrieval, graphically browsing the data, and testing any processing methods. A robust method for performing symbolization is proposed and tested with real data using the programs mentioned above. The method for performing symbolization is based on the use of a median filter bank for extraction of activities within speed of variation ranges. These are used to characterize the signal contents in terms of variability to extract information about short-term activity and long-term trends. The symbolic description of the signal at a given moment is based on this information. The authors´ method also gives information about the credibility of the activity findings and their symbolic representation.<>