Critical elements on fitting the Bayesian multivariate Poisson Lognormal model

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Motivated by a problem on fitting multivariate models to traffic accident data, a detailed discussion of the Multivariate Poisson Lognormal (MPL) model is presented. This paper reveals three critical elements on fitting the MPL model: the setting of initial estimates, hyperparameters and tuning parameters. These issues have not been highlighted in the literature. Based on simulation studies conducted, we have shown that to use the Univariate Poisson Model (UPM) estimates as starting values, at least 20,000 iterations are needed to obtain reliable final estimates. We also illustrated the sensitivity of the specific hyperparameter, which if it is not given extra attention, may affect the final estimates. The last issue is regarding the tuning parameters where they depend on the acceptance rate. Finally, a heuristic algorithm to fit the MPL model is presented. This acts as a guide to ensure that the model works satisfactorily given any data set.

Original languageEnglish
Title of host publication22nd National Symposium on Mathematical Sciences, SKSM 2014: Strengthening Research and Collaboration of Mathematical Sciences in Malaysia
PublisherAmerican Institute of Physics Inc.
Volume1682
ISBN (Electronic)9780735413290
DOIs
Publication statusPublished - 22 Oct 2015
Event22nd National Symposium on Mathematical Sciences: Strengthening Research and Collaboration of Mathematical Sciences in Malaysia, SKSM 2014 - Selangor, Malaysia
Duration: 24 Nov 201426 Nov 2014

Other

Other22nd National Symposium on Mathematical Sciences: Strengthening Research and Collaboration of Mathematical Sciences in Malaysia, SKSM 2014
CountryMalaysia
CitySelangor
Period24/11/1426/11/14

Fingerprint

estimates
tuning
accidents
acceptability
traffic
iteration
sensitivity
simulation

Keywords

  • Bayesian Modelling
  • MCMC
  • Traffic Accident Multivariate data

ASJC Scopus subject areas

  • Physics and Astronomy(all)

Cite this

Zamzuri, Z. H. (2015). Critical elements on fitting the Bayesian multivariate Poisson Lognormal model. In 22nd National Symposium on Mathematical Sciences, SKSM 2014: Strengthening Research and Collaboration of Mathematical Sciences in Malaysia (Vol. 1682). [050005] American Institute of Physics Inc.. https://doi.org/10.1063/1.4932496

Critical elements on fitting the Bayesian multivariate Poisson Lognormal model. / Zamzuri, Zamira Hasanah.

22nd National Symposium on Mathematical Sciences, SKSM 2014: Strengthening Research and Collaboration of Mathematical Sciences in Malaysia. Vol. 1682 American Institute of Physics Inc., 2015. 050005.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Zamzuri, ZH 2015, Critical elements on fitting the Bayesian multivariate Poisson Lognormal model. in 22nd National Symposium on Mathematical Sciences, SKSM 2014: Strengthening Research and Collaboration of Mathematical Sciences in Malaysia. vol. 1682, 050005, American Institute of Physics Inc., 22nd National Symposium on Mathematical Sciences: Strengthening Research and Collaboration of Mathematical Sciences in Malaysia, SKSM 2014, Selangor, Malaysia, 24/11/14. https://doi.org/10.1063/1.4932496
Zamzuri ZH. Critical elements on fitting the Bayesian multivariate Poisson Lognormal model. In 22nd National Symposium on Mathematical Sciences, SKSM 2014: Strengthening Research and Collaboration of Mathematical Sciences in Malaysia. Vol. 1682. American Institute of Physics Inc. 2015. 050005 https://doi.org/10.1063/1.4932496
Zamzuri, Zamira Hasanah. / Critical elements on fitting the Bayesian multivariate Poisson Lognormal model. 22nd National Symposium on Mathematical Sciences, SKSM 2014: Strengthening Research and Collaboration of Mathematical Sciences in Malaysia. Vol. 1682 American Institute of Physics Inc., 2015.
@inproceedings{aafd5de72f284d9bbdb6b5ad6b3716c3,
title = "Critical elements on fitting the Bayesian multivariate Poisson Lognormal model",
abstract = "Motivated by a problem on fitting multivariate models to traffic accident data, a detailed discussion of the Multivariate Poisson Lognormal (MPL) model is presented. This paper reveals three critical elements on fitting the MPL model: the setting of initial estimates, hyperparameters and tuning parameters. These issues have not been highlighted in the literature. Based on simulation studies conducted, we have shown that to use the Univariate Poisson Model (UPM) estimates as starting values, at least 20,000 iterations are needed to obtain reliable final estimates. We also illustrated the sensitivity of the specific hyperparameter, which if it is not given extra attention, may affect the final estimates. The last issue is regarding the tuning parameters where they depend on the acceptance rate. Finally, a heuristic algorithm to fit the MPL model is presented. This acts as a guide to ensure that the model works satisfactorily given any data set.",
keywords = "Bayesian Modelling, MCMC, Traffic Accident Multivariate data",
author = "Zamzuri, {Zamira Hasanah}",
year = "2015",
month = "10",
day = "22",
doi = "10.1063/1.4932496",
language = "English",
volume = "1682",
booktitle = "22nd National Symposium on Mathematical Sciences, SKSM 2014: Strengthening Research and Collaboration of Mathematical Sciences in Malaysia",
publisher = "American Institute of Physics Inc.",

}

TY - GEN

T1 - Critical elements on fitting the Bayesian multivariate Poisson Lognormal model

AU - Zamzuri, Zamira Hasanah

PY - 2015/10/22

Y1 - 2015/10/22

N2 - Motivated by a problem on fitting multivariate models to traffic accident data, a detailed discussion of the Multivariate Poisson Lognormal (MPL) model is presented. This paper reveals three critical elements on fitting the MPL model: the setting of initial estimates, hyperparameters and tuning parameters. These issues have not been highlighted in the literature. Based on simulation studies conducted, we have shown that to use the Univariate Poisson Model (UPM) estimates as starting values, at least 20,000 iterations are needed to obtain reliable final estimates. We also illustrated the sensitivity of the specific hyperparameter, which if it is not given extra attention, may affect the final estimates. The last issue is regarding the tuning parameters where they depend on the acceptance rate. Finally, a heuristic algorithm to fit the MPL model is presented. This acts as a guide to ensure that the model works satisfactorily given any data set.

AB - Motivated by a problem on fitting multivariate models to traffic accident data, a detailed discussion of the Multivariate Poisson Lognormal (MPL) model is presented. This paper reveals three critical elements on fitting the MPL model: the setting of initial estimates, hyperparameters and tuning parameters. These issues have not been highlighted in the literature. Based on simulation studies conducted, we have shown that to use the Univariate Poisson Model (UPM) estimates as starting values, at least 20,000 iterations are needed to obtain reliable final estimates. We also illustrated the sensitivity of the specific hyperparameter, which if it is not given extra attention, may affect the final estimates. The last issue is regarding the tuning parameters where they depend on the acceptance rate. Finally, a heuristic algorithm to fit the MPL model is presented. This acts as a guide to ensure that the model works satisfactorily given any data set.

KW - Bayesian Modelling

KW - MCMC

KW - Traffic Accident Multivariate data

UR - http://www.scopus.com/inward/record.url?scp=84984537137&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84984537137&partnerID=8YFLogxK

U2 - 10.1063/1.4932496

DO - 10.1063/1.4932496

M3 - Conference contribution

VL - 1682

BT - 22nd National Symposium on Mathematical Sciences, SKSM 2014: Strengthening Research and Collaboration of Mathematical Sciences in Malaysia

PB - American Institute of Physics Inc.

ER -