The estimation of the ClusterWeighted Model is particularly attractive for providing explicit modeling of the explanatory variables, in a mixture of regression. The Robust version of the model requires the specification of a set of crucial parameters, like the proportion of trimmed units U, the thresholds to be adopted for the constrained estimation of group scatter and for regression errors, beyond the number of components of the Mixture. To assist the choice of such hyper-parameters, a monitoring methodology could be of great help. The purpose is to provide a set of graphical tools to guide the final user in making an informed judgment, considering a landscape of plausible choices. The final output offers a set of optimal solutions, featured by the interval of hyper-parameters values in which their optimality holds, their stability and validity. An assessment of the role and extent of the outlying observations has been provided, introducing three new silhouette plots. The purpose is to understand the possible effects of the contaminated observations, with respect to the clustering of the covariate -, and the local regression lines, following the nature of the Cluster Weighted model.

Cappozzo, A., García-Escudero, L., Greselin, F., Mayo-Iscar, A. (2022). Monitoring Hyperparameter Choice for Robust Cluster Weighted Model. In Classification and Data Science in the Digital Age - Book of Abstracts IFCS 2022 (pp. 105-105). Printed in Portugal by Instituto Nacional de Estatística.

Monitoring Hyperparameter Choice for Robust Cluster Weighted Model

Andrea Cappozzo;Francesca Greselin
;
2022

Abstract

The estimation of the ClusterWeighted Model is particularly attractive for providing explicit modeling of the explanatory variables, in a mixture of regression. The Robust version of the model requires the specification of a set of crucial parameters, like the proportion of trimmed units U, the thresholds to be adopted for the constrained estimation of group scatter and for regression errors, beyond the number of components of the Mixture. To assist the choice of such hyper-parameters, a monitoring methodology could be of great help. The purpose is to provide a set of graphical tools to guide the final user in making an informed judgment, considering a landscape of plausible choices. The final output offers a set of optimal solutions, featured by the interval of hyper-parameters values in which their optimality holds, their stability and validity. An assessment of the role and extent of the outlying observations has been provided, introducing three new silhouette plots. The purpose is to understand the possible effects of the contaminated observations, with respect to the clustering of the covariate -, and the local regression lines, following the nature of the Cluster Weighted model.
Capitolo o saggio
cluster-weighted modeling, outliers, trimmed bic, eigenvalue constraint, monitoring, model-based clustering, robust estimation
English
Classification and Data Science in the Digital Age - Book of Abstracts IFCS 2022
978-989-98955-9-1
Cappozzo, A., García-Escudero, L., Greselin, F., Mayo-Iscar, A. (2022). Monitoring Hyperparameter Choice for Robust Cluster Weighted Model. In Classification and Data Science in the Digital Age - Book of Abstracts IFCS 2022 (pp. 105-105). Printed in Portugal by Instituto Nacional de Estatística.
Cappozzo, A; García-Escudero, L; Greselin, F; Mayo-Iscar, A
File in questo prodotto:
File Dimensione Formato  
IFCS2022 BoA ISBN CGGM Hyperparameter Choice for Robust Cluster Weighted Model.pdf

accesso aperto

Descrizione: Book of Abstracts
Tipologia di allegato: Publisher’s Version (Version of Record, VoR)
Dimensione 844.37 kB
Formato Adobe PDF
844.37 kB Adobe PDF Visualizza/Apri

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/10281/389289
Citazioni
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
Social impact