Fiche du document
16 juin 2017
- handle: 10670/1.cm1vac
- hal: hal-01656197
- doi: 10.1007/978-3-319-59105-6_14
Ce document est lié à :
info:eu-repo/semantics/altIdentifier/doi/10.1007/978-3-319-59105-6_14
http://creativecommons.org/licenses/by/ , info:eu-repo/semantics/OpenAccess
Mots-clés
Healthy state Abnormal state Non-negative matrix factorization (NMF) Heatmap Self-organizing map (SOM) Inner factors Reconstruction of dataSujets proches
Natural science Sciences Science of science Factors Agents, Manufacturers' Manufacturers' representatives FactorsCiter ce document
Anna M. Bartkowiak et al., « NMF in Screening Some Spirometric Data, an Insight into 12-Dimensional Data Space », HAL-SHS : sciences de l'information, de la communication et des bibliothèques, ID : 10.1007/978-3-319-59105-6_14
Métriques
Partage / Export
Résumé
We present the usage of the Non-negative Matrix Factorization (NMF), an unsupervised machine learning method, which learns normal and abnormal state of patient’s ventilatory systems. This is done using samples of patients having defects of obturative and restrictive kind and a control group.We show that the NMF method can identify patients being in the normal state and screen them off from the remaining patients; however the kind of the ventilatory disorder for the remaining patients is not recognized. This is confronted with clustering provided by the k-means method and visualization of the 12-dimensional data using heatmaps and Kohonen’s self-organizing maps.The data set can be reconstructed with a 0.9746 accuracy (fraction of explained variance) from 6 base vectors provided by the NMF and using appropriate encoders provided also by the NMF; while 3 factors yield an 0.8573 fraction of explained variance.