Development of Methods for Bridging Missing Data Points
During the Identification of FIR Models
Introduction
When recording data streams from biomedical systems, it happens frequently that
some data values are either missing of incorrect (irrelevant). The electrodes
are periodically removed from patients with an ECG monitor to give the nurse
the opportunity of washing the patient. A patient has to sneeze, and as a
consequence thereof, the heart rate changes temporarily.
A methodology for inductive modeling of such systems must be capable
of coping with incomplete data streams and/or data records that contain
outliers.
In this project, the FIR methodology was enhanced by adding precisely such
a feature to the methodology. Algorithms were added that are able to
recognize and eliminate data sets containing either holes or outliers.
Mechanisms were added that enable FIR to continue with an interrupted data
stream, after the interruption has ended.
Finally, schemes were added to correct outliers and/or complete interrupted
data streams. To this end, the data stream is first cleaned by eliminating
data records containing outliers and throwing away incomplete data records.
The so reduced training data are then used to identify a FIR model. The
FIR model is subsequently being used to making "predictions" of the missing
and/or faulty data records.
Most Important Publications
- Nebot, A., and F.E. Cellier (1994),
Preconditioning of Measurement Data for the Elimination of
Patient-Specific Behavior in Qualitative Modeling of Medical
Systems,
Proc. CISS'94, First Joint Conf. of Intl. Simulation Societies,
Zurich, Switzerland, pp.584-588.
- Nebot A., and F.E. Cellier (1994),
Dealing With Incomplete Data Records in Qualitative Modeling and
Simulation of Biomedical Systems,
Proc. CISS'94, First Joint Conf. of Intl. Simulation Societies,
Zurich, Switzerland, pp.605-610.
Sponsors
- Consejo Interministerial de Ciencia y Tecnología (CICYT)
Deutsche Version
Homepage
Last modified: July 18, 2005 -- © François Cellier