Skip to Main content Skip to Navigation

Modèles additifs parcimonieux

Abstract : Many function estimation algorithms exist in supervised statistical learning. However, they have been developed aiming to provide precise estimators, without considering the interpretability of the solution. Additive models allow to explain the predictions simply, dealing with only one explanatory variable at a time, but they are difficult to implement. This thesis develops an estimation algorithm for additive models. On the one hand, their use is simplified, since the complexity tuning is mainly integrated in the parameter estimation phase. On the other hand, the interpretability is also supported by a tendency to automatically eliminate the least relevant variables. Strategies for accelerating computation are also proposed. An approximation of the effective number of parameters allows the use of model selection analytical criteria. Its validity is tested by simulations and on real data.
Document type :
Complete list of metadatas

Cited literature [154 references]  Display  Hide  Download
Contributor : Marta Avalos <>
Submitted on : Wednesday, March 16, 2005 - 6:24:44 PM
Last modification on : Monday, December 21, 2020 - 5:26:03 PM
Long-term archiving on: : Friday, April 2, 2010 - 9:20:47 PM


  • HAL Id : tel-00008802, version 1



Marta Avalos. Modèles additifs parcimonieux. Autre [cs.OH]. Université de Technologie de Compiègne, 2004. Français. ⟨tel-00008802⟩



Record views


Files downloads