This work proposes a sequential procedure to select the best model among several nested non-linear models and to estimate efficiently the parameters of the chosen model. At the first step of this procedure, a generalized DKL-optimum design is computed that is optimal for the goals of model selection and parameter estimation. Subsequently, at each step, an adaptive generalized DKL-optimum design is computed from the data accrued and the tests previously performed. The proposed sequential scheme selects the best non-linear model with probability converging to one; moreover it allows efficient estimates of parameters, since the adaptive sequential DKL-optimum designs converge to the D-optimum design for the "true" model.
Model selection and parameter estimation in non-linear nested models : a sequential generalized DKL-optimum design / C. Tommasi, C. May. - In: STATISTICA SINICA. - ISSN 1017-0405. - 24:1(2014), pp. 63-82.
Model selection and parameter estimation in non-linear nested models : a sequential generalized DKL-optimum design
C. TommasiUltimo
;
2014
Abstract
This work proposes a sequential procedure to select the best model among several nested non-linear models and to estimate efficiently the parameters of the chosen model. At the first step of this procedure, a generalized DKL-optimum design is computed that is optimal for the goals of model selection and parameter estimation. Subsequently, at each step, an adaptive generalized DKL-optimum design is computed from the data accrued and the tests previously performed. The proposed sequential scheme selects the best non-linear model with probability converging to one; moreover it allows efficient estimates of parameters, since the adaptive sequential DKL-optimum designs converge to the D-optimum design for the "true" model.File | Dimensione | Formato | |
---|---|---|---|
may-tommasi.pdf
accesso riservato
Tipologia:
Publisher's version/PDF
Dimensione
188.42 kB
Formato
Adobe PDF
|
188.42 kB | Adobe PDF | Visualizza/Apri Richiedi una copia |
Pubblicazioni consigliate
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.