|
Elisa Fromont and Marie-Odile Cordier and René Quiniou
Extraction de connaissances provenant de données multisources pour la caractérisation d'arythmies cardiaques
, Chapitre Fouille de données complexes
, Vol. RNTI-E-4
, 25-45
, 2005
|
|
Abstract
In many applications dealing with industrial or medical supervision,
data are temporal time series related to numerous sensors that provide
information which is complementary but also often redundant. We
investigate the problem of learning, by inductive logic programming,
symbolic rules that characterize cardiac arrhythmias from multisource
data such as electrocardiograms or arterial blood pressure measures. A
first strategy consists in aggregating the data and then in learning
directly from these transformed data. This method is not very
efficient and it is difficult to implement, especially designing the
learning bias, when the amount of data is big. We propose an efficient
two-step strategy that uses monosource learning to automatically bias
and reduce the search space for multisource learning. The results
obtained with this method are analyzed and compared to those obtained
with the naive learning method. We show that an order of magnitude is
gained on learning times with the new method.
|
|