Skip to Main content Skip to Navigation

Apprentissage multisource par programmation logique inductive : application à la caractérisation d'arythmies cardiaques

Elisa Fromont 1
1 DREAM - Diagnosing, Recommending Actions and Modelling
Abstract : This work belongs to the field of knowledge discovery from data coming from several sources that describe a sole phenomenon. The aim is to improve the quality of monitoring systems. When data are redundant, several sources can help the system to cope with the loss of a signal or the presence of important noise. When sources are complementary, the joint use of all the data coming from the different sources can improve the detection performances of the systems. Our work is applied to cardiac arrhythmia diagnosis. We use a relational learning technique (inductive logic programming) to learn discriminating rules that characterize cardiac arrhythmias from different leads of an electrocardiogram and a measure of arterial blood pressure. To exploit redundant sources, we designed appropriate solutions to learn accurate monosource rules from data coming from each source separately. If the sources are complementary, a naive multisource learning method that consists in aggregating the data of all the sources and then learn globally from the whole set of data with a learning bias as few restrictive as possible, is a first solution to produce rules that contain relationships between events occurring on the different sources. However, this technique is not efficient and the huge search space associated with the set of possible solutions for the multisource problem leads to unsatisfactory solutions when using a relational learner. To solve those dimensionality problems, we have proposed a new multisource learning method that uses monosource rules to bias automatically and efficiently a new learning process from the aggregated data. The work has been done in the context of the Cepica project. The results show that the proposed method is much more efficient than learning directly from the aggregated data. Furthermore, it yields rules having better or equal accuracy than rules obtained by monosource learning. Besides, the learned rules have been transformed into chronicles and implemented in the cardiac monitoring system Calicot and tested with encouraging results on real noisy signal.
Complete list of metadatas

Cited literature [126 references]  Display  Hide  Download
Contributor : Elisa Fromont <>
Submitted on : Monday, February 13, 2006 - 10:04:59 AM
Last modification on : Monday, October 19, 2020 - 11:07:31 AM
Long-term archiving on: : Monday, September 20, 2010 - 1:31:15 PM


  • HAL Id : tel-00011455, version 2


Elisa Fromont. Apprentissage multisource par programmation logique inductive : application à la caractérisation d'arythmies cardiaques. Autre [cs.OH]. Université Rennes 1, 2005. Français. ⟨tel-00011455v2⟩



Record views


Files downloads