Comparison of Bayesian predictive methods for model selection
Abstract The goal of this paper is to compare several widely used Bayesian model selection methods in practical model selection problems, highlight their differences and give recommendations about the preferred approaches. We focus on the variable subset selection for regression and classification a...
Ausführliche Beschreibung
Autor*in: |
Piironen, Juho [verfasserIn] |
---|
Format: |
Artikel |
---|---|
Sprache: |
Englisch |
Erschienen: |
2016 |
---|
Schlagwörter: |
---|
Anmerkung: |
© The Author(s) 2016 |
---|
Übergeordnetes Werk: |
Enthalten in: Statistics and computing - Springer US, 1991, 27(2016), 3 vom: 07. Apr., Seite 711-735 |
---|---|
Übergeordnetes Werk: |
volume:27 ; year:2016 ; number:3 ; day:07 ; month:04 ; pages:711-735 |
Links: |
---|
DOI / URN: |
10.1007/s11222-016-9649-y |
---|
Katalog-ID: |
OLC2033748898 |
---|
LEADER | 01000caa a22002652 4500 | ||
---|---|---|---|
001 | OLC2033748898 | ||
003 | DE-627 | ||
005 | 20230504051513.0 | ||
007 | tu | ||
008 | 200819s2016 xx ||||| 00| ||eng c | ||
024 | 7 | |a 10.1007/s11222-016-9649-y |2 doi | |
035 | |a (DE-627)OLC2033748898 | ||
035 | |a (DE-He213)s11222-016-9649-y-p | ||
040 | |a DE-627 |b ger |c DE-627 |e rakwb | ||
041 | |a eng | ||
082 | 0 | 4 | |a 004 |a 620 |q VZ |
100 | 1 | |a Piironen, Juho |e verfasserin |4 aut | |
245 | 1 | 0 | |a Comparison of Bayesian predictive methods for model selection |
264 | 1 | |c 2016 | |
336 | |a Text |b txt |2 rdacontent | ||
337 | |a ohne Hilfsmittel zu benutzen |b n |2 rdamedia | ||
338 | |a Band |b nc |2 rdacarrier | ||
500 | |a © The Author(s) 2016 | ||
520 | |a Abstract The goal of this paper is to compare several widely used Bayesian model selection methods in practical model selection problems, highlight their differences and give recommendations about the preferred approaches. We focus on the variable subset selection for regression and classification and perform several numerical experiments using both simulated and real world data. The results show that the optimization of a utility estimate such as the cross-validation (CV) score is liable to finding overfitted models due to relatively high variance in the utility estimates when the data is scarce. This can also lead to substantial selection induced bias and optimism in the performance evaluation for the selected model. From a predictive viewpoint, best results are obtained by accounting for model uncertainty by forming the full encompassing model, such as the Bayesian model averaging solution over the candidate models. If the encompassing model is too complex, it can be robustly simplified by the projection method, in which the information of the full model is projected onto the submodels. This approach is substantially less prone to overfitting than selection based on CV-score. Overall, the projection method appears to outperform also the maximum a posteriori model and the selection of the most probable variables. The study also demonstrates that the model selection can greatly benefit from using cross-validation outside the searching process both for guiding the model size selection and assessing the predictive performance of the finally selected model. | ||
650 | 4 | |a Bayesian model selection | |
650 | 4 | |a Cross-validation | |
650 | 4 | |a Reference model | |
650 | 4 | |a Projection | |
650 | 4 | |a Selection bias | |
700 | 1 | |a Vehtari, Aki |4 aut | |
773 | 0 | 8 | |i Enthalten in |t Statistics and computing |d Springer US, 1991 |g 27(2016), 3 vom: 07. Apr., Seite 711-735 |w (DE-627)131007963 |w (DE-600)1087487-2 |w (DE-576)052732894 |x 0960-3174 |7 nnns |
773 | 1 | 8 | |g volume:27 |g year:2016 |g number:3 |g day:07 |g month:04 |g pages:711-735 |
856 | 4 | 1 | |u https://doi.org/10.1007/s11222-016-9649-y |z lizenzpflichtig |3 Volltext |
912 | |a GBV_USEFLAG_A | ||
912 | |a SYSFLAG_A | ||
912 | |a GBV_OLC | ||
912 | |a SSG-OLC-TEC | ||
912 | |a SSG-OLC-MAT | ||
912 | |a GBV_ILN_70 | ||
912 | |a GBV_ILN_4126 | ||
951 | |a AR | ||
952 | |d 27 |j 2016 |e 3 |b 07 |c 04 |h 711-735 |
author_variant |
j p jp a v av |
---|---|
matchkey_str |
article:09603174:2016----::oprsnfaeinrdcieehdf |
hierarchy_sort_str |
2016 |
publishDate |
2016 |
allfields |
10.1007/s11222-016-9649-y doi (DE-627)OLC2033748898 (DE-He213)s11222-016-9649-y-p DE-627 ger DE-627 rakwb eng 004 620 VZ Piironen, Juho verfasserin aut Comparison of Bayesian predictive methods for model selection 2016 Text txt rdacontent ohne Hilfsmittel zu benutzen n rdamedia Band nc rdacarrier © The Author(s) 2016 Abstract The goal of this paper is to compare several widely used Bayesian model selection methods in practical model selection problems, highlight their differences and give recommendations about the preferred approaches. We focus on the variable subset selection for regression and classification and perform several numerical experiments using both simulated and real world data. The results show that the optimization of a utility estimate such as the cross-validation (CV) score is liable to finding overfitted models due to relatively high variance in the utility estimates when the data is scarce. This can also lead to substantial selection induced bias and optimism in the performance evaluation for the selected model. From a predictive viewpoint, best results are obtained by accounting for model uncertainty by forming the full encompassing model, such as the Bayesian model averaging solution over the candidate models. If the encompassing model is too complex, it can be robustly simplified by the projection method, in which the information of the full model is projected onto the submodels. This approach is substantially less prone to overfitting than selection based on CV-score. Overall, the projection method appears to outperform also the maximum a posteriori model and the selection of the most probable variables. The study also demonstrates that the model selection can greatly benefit from using cross-validation outside the searching process both for guiding the model size selection and assessing the predictive performance of the finally selected model. Bayesian model selection Cross-validation Reference model Projection Selection bias Vehtari, Aki aut Enthalten in Statistics and computing Springer US, 1991 27(2016), 3 vom: 07. Apr., Seite 711-735 (DE-627)131007963 (DE-600)1087487-2 (DE-576)052732894 0960-3174 nnns volume:27 year:2016 number:3 day:07 month:04 pages:711-735 https://doi.org/10.1007/s11222-016-9649-y lizenzpflichtig Volltext GBV_USEFLAG_A SYSFLAG_A GBV_OLC SSG-OLC-TEC SSG-OLC-MAT GBV_ILN_70 GBV_ILN_4126 AR 27 2016 3 07 04 711-735 |
spelling |
10.1007/s11222-016-9649-y doi (DE-627)OLC2033748898 (DE-He213)s11222-016-9649-y-p DE-627 ger DE-627 rakwb eng 004 620 VZ Piironen, Juho verfasserin aut Comparison of Bayesian predictive methods for model selection 2016 Text txt rdacontent ohne Hilfsmittel zu benutzen n rdamedia Band nc rdacarrier © The Author(s) 2016 Abstract The goal of this paper is to compare several widely used Bayesian model selection methods in practical model selection problems, highlight their differences and give recommendations about the preferred approaches. We focus on the variable subset selection for regression and classification and perform several numerical experiments using both simulated and real world data. The results show that the optimization of a utility estimate such as the cross-validation (CV) score is liable to finding overfitted models due to relatively high variance in the utility estimates when the data is scarce. This can also lead to substantial selection induced bias and optimism in the performance evaluation for the selected model. From a predictive viewpoint, best results are obtained by accounting for model uncertainty by forming the full encompassing model, such as the Bayesian model averaging solution over the candidate models. If the encompassing model is too complex, it can be robustly simplified by the projection method, in which the information of the full model is projected onto the submodels. This approach is substantially less prone to overfitting than selection based on CV-score. Overall, the projection method appears to outperform also the maximum a posteriori model and the selection of the most probable variables. The study also demonstrates that the model selection can greatly benefit from using cross-validation outside the searching process both for guiding the model size selection and assessing the predictive performance of the finally selected model. Bayesian model selection Cross-validation Reference model Projection Selection bias Vehtari, Aki aut Enthalten in Statistics and computing Springer US, 1991 27(2016), 3 vom: 07. Apr., Seite 711-735 (DE-627)131007963 (DE-600)1087487-2 (DE-576)052732894 0960-3174 nnns volume:27 year:2016 number:3 day:07 month:04 pages:711-735 https://doi.org/10.1007/s11222-016-9649-y lizenzpflichtig Volltext GBV_USEFLAG_A SYSFLAG_A GBV_OLC SSG-OLC-TEC SSG-OLC-MAT GBV_ILN_70 GBV_ILN_4126 AR 27 2016 3 07 04 711-735 |
allfields_unstemmed |
10.1007/s11222-016-9649-y doi (DE-627)OLC2033748898 (DE-He213)s11222-016-9649-y-p DE-627 ger DE-627 rakwb eng 004 620 VZ Piironen, Juho verfasserin aut Comparison of Bayesian predictive methods for model selection 2016 Text txt rdacontent ohne Hilfsmittel zu benutzen n rdamedia Band nc rdacarrier © The Author(s) 2016 Abstract The goal of this paper is to compare several widely used Bayesian model selection methods in practical model selection problems, highlight their differences and give recommendations about the preferred approaches. We focus on the variable subset selection for regression and classification and perform several numerical experiments using both simulated and real world data. The results show that the optimization of a utility estimate such as the cross-validation (CV) score is liable to finding overfitted models due to relatively high variance in the utility estimates when the data is scarce. This can also lead to substantial selection induced bias and optimism in the performance evaluation for the selected model. From a predictive viewpoint, best results are obtained by accounting for model uncertainty by forming the full encompassing model, such as the Bayesian model averaging solution over the candidate models. If the encompassing model is too complex, it can be robustly simplified by the projection method, in which the information of the full model is projected onto the submodels. This approach is substantially less prone to overfitting than selection based on CV-score. Overall, the projection method appears to outperform also the maximum a posteriori model and the selection of the most probable variables. The study also demonstrates that the model selection can greatly benefit from using cross-validation outside the searching process both for guiding the model size selection and assessing the predictive performance of the finally selected model. Bayesian model selection Cross-validation Reference model Projection Selection bias Vehtari, Aki aut Enthalten in Statistics and computing Springer US, 1991 27(2016), 3 vom: 07. Apr., Seite 711-735 (DE-627)131007963 (DE-600)1087487-2 (DE-576)052732894 0960-3174 nnns volume:27 year:2016 number:3 day:07 month:04 pages:711-735 https://doi.org/10.1007/s11222-016-9649-y lizenzpflichtig Volltext GBV_USEFLAG_A SYSFLAG_A GBV_OLC SSG-OLC-TEC SSG-OLC-MAT GBV_ILN_70 GBV_ILN_4126 AR 27 2016 3 07 04 711-735 |
allfieldsGer |
10.1007/s11222-016-9649-y doi (DE-627)OLC2033748898 (DE-He213)s11222-016-9649-y-p DE-627 ger DE-627 rakwb eng 004 620 VZ Piironen, Juho verfasserin aut Comparison of Bayesian predictive methods for model selection 2016 Text txt rdacontent ohne Hilfsmittel zu benutzen n rdamedia Band nc rdacarrier © The Author(s) 2016 Abstract The goal of this paper is to compare several widely used Bayesian model selection methods in practical model selection problems, highlight their differences and give recommendations about the preferred approaches. We focus on the variable subset selection for regression and classification and perform several numerical experiments using both simulated and real world data. The results show that the optimization of a utility estimate such as the cross-validation (CV) score is liable to finding overfitted models due to relatively high variance in the utility estimates when the data is scarce. This can also lead to substantial selection induced bias and optimism in the performance evaluation for the selected model. From a predictive viewpoint, best results are obtained by accounting for model uncertainty by forming the full encompassing model, such as the Bayesian model averaging solution over the candidate models. If the encompassing model is too complex, it can be robustly simplified by the projection method, in which the information of the full model is projected onto the submodels. This approach is substantially less prone to overfitting than selection based on CV-score. Overall, the projection method appears to outperform also the maximum a posteriori model and the selection of the most probable variables. The study also demonstrates that the model selection can greatly benefit from using cross-validation outside the searching process both for guiding the model size selection and assessing the predictive performance of the finally selected model. Bayesian model selection Cross-validation Reference model Projection Selection bias Vehtari, Aki aut Enthalten in Statistics and computing Springer US, 1991 27(2016), 3 vom: 07. Apr., Seite 711-735 (DE-627)131007963 (DE-600)1087487-2 (DE-576)052732894 0960-3174 nnns volume:27 year:2016 number:3 day:07 month:04 pages:711-735 https://doi.org/10.1007/s11222-016-9649-y lizenzpflichtig Volltext GBV_USEFLAG_A SYSFLAG_A GBV_OLC SSG-OLC-TEC SSG-OLC-MAT GBV_ILN_70 GBV_ILN_4126 AR 27 2016 3 07 04 711-735 |
allfieldsSound |
10.1007/s11222-016-9649-y doi (DE-627)OLC2033748898 (DE-He213)s11222-016-9649-y-p DE-627 ger DE-627 rakwb eng 004 620 VZ Piironen, Juho verfasserin aut Comparison of Bayesian predictive methods for model selection 2016 Text txt rdacontent ohne Hilfsmittel zu benutzen n rdamedia Band nc rdacarrier © The Author(s) 2016 Abstract The goal of this paper is to compare several widely used Bayesian model selection methods in practical model selection problems, highlight their differences and give recommendations about the preferred approaches. We focus on the variable subset selection for regression and classification and perform several numerical experiments using both simulated and real world data. The results show that the optimization of a utility estimate such as the cross-validation (CV) score is liable to finding overfitted models due to relatively high variance in the utility estimates when the data is scarce. This can also lead to substantial selection induced bias and optimism in the performance evaluation for the selected model. From a predictive viewpoint, best results are obtained by accounting for model uncertainty by forming the full encompassing model, such as the Bayesian model averaging solution over the candidate models. If the encompassing model is too complex, it can be robustly simplified by the projection method, in which the information of the full model is projected onto the submodels. This approach is substantially less prone to overfitting than selection based on CV-score. Overall, the projection method appears to outperform also the maximum a posteriori model and the selection of the most probable variables. The study also demonstrates that the model selection can greatly benefit from using cross-validation outside the searching process both for guiding the model size selection and assessing the predictive performance of the finally selected model. Bayesian model selection Cross-validation Reference model Projection Selection bias Vehtari, Aki aut Enthalten in Statistics and computing Springer US, 1991 27(2016), 3 vom: 07. Apr., Seite 711-735 (DE-627)131007963 (DE-600)1087487-2 (DE-576)052732894 0960-3174 nnns volume:27 year:2016 number:3 day:07 month:04 pages:711-735 https://doi.org/10.1007/s11222-016-9649-y lizenzpflichtig Volltext GBV_USEFLAG_A SYSFLAG_A GBV_OLC SSG-OLC-TEC SSG-OLC-MAT GBV_ILN_70 GBV_ILN_4126 AR 27 2016 3 07 04 711-735 |
language |
English |
source |
Enthalten in Statistics and computing 27(2016), 3 vom: 07. Apr., Seite 711-735 volume:27 year:2016 number:3 day:07 month:04 pages:711-735 |
sourceStr |
Enthalten in Statistics and computing 27(2016), 3 vom: 07. Apr., Seite 711-735 volume:27 year:2016 number:3 day:07 month:04 pages:711-735 |
format_phy_str_mv |
Article |
institution |
findex.gbv.de |
topic_facet |
Bayesian model selection Cross-validation Reference model Projection Selection bias |
dewey-raw |
004 |
isfreeaccess_bool |
false |
container_title |
Statistics and computing |
authorswithroles_txt_mv |
Piironen, Juho @@aut@@ Vehtari, Aki @@aut@@ |
publishDateDaySort_date |
2016-04-07T00:00:00Z |
hierarchy_top_id |
131007963 |
dewey-sort |
14 |
id |
OLC2033748898 |
language_de |
englisch |
fullrecord |
<?xml version="1.0" encoding="UTF-8"?><collection xmlns="http://www.loc.gov/MARC21/slim"><record><leader>01000caa a22002652 4500</leader><controlfield tag="001">OLC2033748898</controlfield><controlfield tag="003">DE-627</controlfield><controlfield tag="005">20230504051513.0</controlfield><controlfield tag="007">tu</controlfield><controlfield tag="008">200819s2016 xx ||||| 00| ||eng c</controlfield><datafield tag="024" ind1="7" ind2=" "><subfield code="a">10.1007/s11222-016-9649-y</subfield><subfield code="2">doi</subfield></datafield><datafield tag="035" ind1=" " ind2=" "><subfield code="a">(DE-627)OLC2033748898</subfield></datafield><datafield tag="035" ind1=" " ind2=" "><subfield code="a">(DE-He213)s11222-016-9649-y-p</subfield></datafield><datafield tag="040" ind1=" " ind2=" "><subfield code="a">DE-627</subfield><subfield code="b">ger</subfield><subfield code="c">DE-627</subfield><subfield code="e">rakwb</subfield></datafield><datafield tag="041" ind1=" " ind2=" "><subfield code="a">eng</subfield></datafield><datafield tag="082" ind1="0" ind2="4"><subfield code="a">004</subfield><subfield code="a">620</subfield><subfield code="q">VZ</subfield></datafield><datafield tag="100" ind1="1" ind2=" "><subfield code="a">Piironen, Juho</subfield><subfield code="e">verfasserin</subfield><subfield code="4">aut</subfield></datafield><datafield tag="245" ind1="1" ind2="0"><subfield code="a">Comparison of Bayesian predictive methods for model selection</subfield></datafield><datafield tag="264" ind1=" " ind2="1"><subfield code="c">2016</subfield></datafield><datafield tag="336" ind1=" " ind2=" "><subfield code="a">Text</subfield><subfield code="b">txt</subfield><subfield code="2">rdacontent</subfield></datafield><datafield tag="337" ind1=" " ind2=" "><subfield code="a">ohne Hilfsmittel zu benutzen</subfield><subfield code="b">n</subfield><subfield code="2">rdamedia</subfield></datafield><datafield tag="338" ind1=" " ind2=" "><subfield code="a">Band</subfield><subfield code="b">nc</subfield><subfield code="2">rdacarrier</subfield></datafield><datafield tag="500" ind1=" " ind2=" "><subfield code="a">© The Author(s) 2016</subfield></datafield><datafield tag="520" ind1=" " ind2=" "><subfield code="a">Abstract The goal of this paper is to compare several widely used Bayesian model selection methods in practical model selection problems, highlight their differences and give recommendations about the preferred approaches. We focus on the variable subset selection for regression and classification and perform several numerical experiments using both simulated and real world data. The results show that the optimization of a utility estimate such as the cross-validation (CV) score is liable to finding overfitted models due to relatively high variance in the utility estimates when the data is scarce. This can also lead to substantial selection induced bias and optimism in the performance evaluation for the selected model. From a predictive viewpoint, best results are obtained by accounting for model uncertainty by forming the full encompassing model, such as the Bayesian model averaging solution over the candidate models. If the encompassing model is too complex, it can be robustly simplified by the projection method, in which the information of the full model is projected onto the submodels. This approach is substantially less prone to overfitting than selection based on CV-score. Overall, the projection method appears to outperform also the maximum a posteriori model and the selection of the most probable variables. The study also demonstrates that the model selection can greatly benefit from using cross-validation outside the searching process both for guiding the model size selection and assessing the predictive performance of the finally selected model.</subfield></datafield><datafield tag="650" ind1=" " ind2="4"><subfield code="a">Bayesian model selection</subfield></datafield><datafield tag="650" ind1=" " ind2="4"><subfield code="a">Cross-validation</subfield></datafield><datafield tag="650" ind1=" " ind2="4"><subfield code="a">Reference model</subfield></datafield><datafield tag="650" ind1=" " ind2="4"><subfield code="a">Projection</subfield></datafield><datafield tag="650" ind1=" " ind2="4"><subfield code="a">Selection bias</subfield></datafield><datafield tag="700" ind1="1" ind2=" "><subfield code="a">Vehtari, Aki</subfield><subfield code="4">aut</subfield></datafield><datafield tag="773" ind1="0" ind2="8"><subfield code="i">Enthalten in</subfield><subfield code="t">Statistics and computing</subfield><subfield code="d">Springer US, 1991</subfield><subfield code="g">27(2016), 3 vom: 07. Apr., Seite 711-735</subfield><subfield code="w">(DE-627)131007963</subfield><subfield code="w">(DE-600)1087487-2</subfield><subfield code="w">(DE-576)052732894</subfield><subfield code="x">0960-3174</subfield><subfield code="7">nnns</subfield></datafield><datafield tag="773" ind1="1" ind2="8"><subfield code="g">volume:27</subfield><subfield code="g">year:2016</subfield><subfield code="g">number:3</subfield><subfield code="g">day:07</subfield><subfield code="g">month:04</subfield><subfield code="g">pages:711-735</subfield></datafield><datafield tag="856" ind1="4" ind2="1"><subfield code="u">https://doi.org/10.1007/s11222-016-9649-y</subfield><subfield code="z">lizenzpflichtig</subfield><subfield code="3">Volltext</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_USEFLAG_A</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">SYSFLAG_A</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_OLC</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">SSG-OLC-TEC</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">SSG-OLC-MAT</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_ILN_70</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_ILN_4126</subfield></datafield><datafield tag="951" ind1=" " ind2=" "><subfield code="a">AR</subfield></datafield><datafield tag="952" ind1=" " ind2=" "><subfield code="d">27</subfield><subfield code="j">2016</subfield><subfield code="e">3</subfield><subfield code="b">07</subfield><subfield code="c">04</subfield><subfield code="h">711-735</subfield></datafield></record></collection>
|
author |
Piironen, Juho |
spellingShingle |
Piironen, Juho ddc 004 misc Bayesian model selection misc Cross-validation misc Reference model misc Projection misc Selection bias Comparison of Bayesian predictive methods for model selection |
authorStr |
Piironen, Juho |
ppnlink_with_tag_str_mv |
@@773@@(DE-627)131007963 |
format |
Article |
dewey-ones |
004 - Data processing & computer science 620 - Engineering & allied operations |
delete_txt_mv |
keep |
author_role |
aut aut |
collection |
OLC |
remote_str |
false |
illustrated |
Not Illustrated |
issn |
0960-3174 |
topic_title |
004 620 VZ Comparison of Bayesian predictive methods for model selection Bayesian model selection Cross-validation Reference model Projection Selection bias |
topic |
ddc 004 misc Bayesian model selection misc Cross-validation misc Reference model misc Projection misc Selection bias |
topic_unstemmed |
ddc 004 misc Bayesian model selection misc Cross-validation misc Reference model misc Projection misc Selection bias |
topic_browse |
ddc 004 misc Bayesian model selection misc Cross-validation misc Reference model misc Projection misc Selection bias |
format_facet |
Aufsätze Gedruckte Aufsätze |
format_main_str_mv |
Text Zeitschrift/Artikel |
carriertype_str_mv |
nc |
hierarchy_parent_title |
Statistics and computing |
hierarchy_parent_id |
131007963 |
dewey-tens |
000 - Computer science, knowledge & systems 620 - Engineering |
hierarchy_top_title |
Statistics and computing |
isfreeaccess_txt |
false |
familylinks_str_mv |
(DE-627)131007963 (DE-600)1087487-2 (DE-576)052732894 |
title |
Comparison of Bayesian predictive methods for model selection |
ctrlnum |
(DE-627)OLC2033748898 (DE-He213)s11222-016-9649-y-p |
title_full |
Comparison of Bayesian predictive methods for model selection |
author_sort |
Piironen, Juho |
journal |
Statistics and computing |
journalStr |
Statistics and computing |
lang_code |
eng |
isOA_bool |
false |
dewey-hundreds |
000 - Computer science, information & general works 600 - Technology |
recordtype |
marc |
publishDateSort |
2016 |
contenttype_str_mv |
txt |
container_start_page |
711 |
author_browse |
Piironen, Juho Vehtari, Aki |
container_volume |
27 |
class |
004 620 VZ |
format_se |
Aufsätze |
author-letter |
Piironen, Juho |
doi_str_mv |
10.1007/s11222-016-9649-y |
dewey-full |
004 620 |
title_sort |
comparison of bayesian predictive methods for model selection |
title_auth |
Comparison of Bayesian predictive methods for model selection |
abstract |
Abstract The goal of this paper is to compare several widely used Bayesian model selection methods in practical model selection problems, highlight their differences and give recommendations about the preferred approaches. We focus on the variable subset selection for regression and classification and perform several numerical experiments using both simulated and real world data. The results show that the optimization of a utility estimate such as the cross-validation (CV) score is liable to finding overfitted models due to relatively high variance in the utility estimates when the data is scarce. This can also lead to substantial selection induced bias and optimism in the performance evaluation for the selected model. From a predictive viewpoint, best results are obtained by accounting for model uncertainty by forming the full encompassing model, such as the Bayesian model averaging solution over the candidate models. If the encompassing model is too complex, it can be robustly simplified by the projection method, in which the information of the full model is projected onto the submodels. This approach is substantially less prone to overfitting than selection based on CV-score. Overall, the projection method appears to outperform also the maximum a posteriori model and the selection of the most probable variables. The study also demonstrates that the model selection can greatly benefit from using cross-validation outside the searching process both for guiding the model size selection and assessing the predictive performance of the finally selected model. © The Author(s) 2016 |
abstractGer |
Abstract The goal of this paper is to compare several widely used Bayesian model selection methods in practical model selection problems, highlight their differences and give recommendations about the preferred approaches. We focus on the variable subset selection for regression and classification and perform several numerical experiments using both simulated and real world data. The results show that the optimization of a utility estimate such as the cross-validation (CV) score is liable to finding overfitted models due to relatively high variance in the utility estimates when the data is scarce. This can also lead to substantial selection induced bias and optimism in the performance evaluation for the selected model. From a predictive viewpoint, best results are obtained by accounting for model uncertainty by forming the full encompassing model, such as the Bayesian model averaging solution over the candidate models. If the encompassing model is too complex, it can be robustly simplified by the projection method, in which the information of the full model is projected onto the submodels. This approach is substantially less prone to overfitting than selection based on CV-score. Overall, the projection method appears to outperform also the maximum a posteriori model and the selection of the most probable variables. The study also demonstrates that the model selection can greatly benefit from using cross-validation outside the searching process both for guiding the model size selection and assessing the predictive performance of the finally selected model. © The Author(s) 2016 |
abstract_unstemmed |
Abstract The goal of this paper is to compare several widely used Bayesian model selection methods in practical model selection problems, highlight their differences and give recommendations about the preferred approaches. We focus on the variable subset selection for regression and classification and perform several numerical experiments using both simulated and real world data. The results show that the optimization of a utility estimate such as the cross-validation (CV) score is liable to finding overfitted models due to relatively high variance in the utility estimates when the data is scarce. This can also lead to substantial selection induced bias and optimism in the performance evaluation for the selected model. From a predictive viewpoint, best results are obtained by accounting for model uncertainty by forming the full encompassing model, such as the Bayesian model averaging solution over the candidate models. If the encompassing model is too complex, it can be robustly simplified by the projection method, in which the information of the full model is projected onto the submodels. This approach is substantially less prone to overfitting than selection based on CV-score. Overall, the projection method appears to outperform also the maximum a posteriori model and the selection of the most probable variables. The study also demonstrates that the model selection can greatly benefit from using cross-validation outside the searching process both for guiding the model size selection and assessing the predictive performance of the finally selected model. © The Author(s) 2016 |
collection_details |
GBV_USEFLAG_A SYSFLAG_A GBV_OLC SSG-OLC-TEC SSG-OLC-MAT GBV_ILN_70 GBV_ILN_4126 |
container_issue |
3 |
title_short |
Comparison of Bayesian predictive methods for model selection |
url |
https://doi.org/10.1007/s11222-016-9649-y |
remote_bool |
false |
author2 |
Vehtari, Aki |
author2Str |
Vehtari, Aki |
ppnlink |
131007963 |
mediatype_str_mv |
n |
isOA_txt |
false |
hochschulschrift_bool |
false |
doi_str |
10.1007/s11222-016-9649-y |
up_date |
2024-07-03T18:17:29.634Z |
_version_ |
1803582858473766912 |
fullrecord_marcxml |
<?xml version="1.0" encoding="UTF-8"?><collection xmlns="http://www.loc.gov/MARC21/slim"><record><leader>01000caa a22002652 4500</leader><controlfield tag="001">OLC2033748898</controlfield><controlfield tag="003">DE-627</controlfield><controlfield tag="005">20230504051513.0</controlfield><controlfield tag="007">tu</controlfield><controlfield tag="008">200819s2016 xx ||||| 00| ||eng c</controlfield><datafield tag="024" ind1="7" ind2=" "><subfield code="a">10.1007/s11222-016-9649-y</subfield><subfield code="2">doi</subfield></datafield><datafield tag="035" ind1=" " ind2=" "><subfield code="a">(DE-627)OLC2033748898</subfield></datafield><datafield tag="035" ind1=" " ind2=" "><subfield code="a">(DE-He213)s11222-016-9649-y-p</subfield></datafield><datafield tag="040" ind1=" " ind2=" "><subfield code="a">DE-627</subfield><subfield code="b">ger</subfield><subfield code="c">DE-627</subfield><subfield code="e">rakwb</subfield></datafield><datafield tag="041" ind1=" " ind2=" "><subfield code="a">eng</subfield></datafield><datafield tag="082" ind1="0" ind2="4"><subfield code="a">004</subfield><subfield code="a">620</subfield><subfield code="q">VZ</subfield></datafield><datafield tag="100" ind1="1" ind2=" "><subfield code="a">Piironen, Juho</subfield><subfield code="e">verfasserin</subfield><subfield code="4">aut</subfield></datafield><datafield tag="245" ind1="1" ind2="0"><subfield code="a">Comparison of Bayesian predictive methods for model selection</subfield></datafield><datafield tag="264" ind1=" " ind2="1"><subfield code="c">2016</subfield></datafield><datafield tag="336" ind1=" " ind2=" "><subfield code="a">Text</subfield><subfield code="b">txt</subfield><subfield code="2">rdacontent</subfield></datafield><datafield tag="337" ind1=" " ind2=" "><subfield code="a">ohne Hilfsmittel zu benutzen</subfield><subfield code="b">n</subfield><subfield code="2">rdamedia</subfield></datafield><datafield tag="338" ind1=" " ind2=" "><subfield code="a">Band</subfield><subfield code="b">nc</subfield><subfield code="2">rdacarrier</subfield></datafield><datafield tag="500" ind1=" " ind2=" "><subfield code="a">© The Author(s) 2016</subfield></datafield><datafield tag="520" ind1=" " ind2=" "><subfield code="a">Abstract The goal of this paper is to compare several widely used Bayesian model selection methods in practical model selection problems, highlight their differences and give recommendations about the preferred approaches. We focus on the variable subset selection for regression and classification and perform several numerical experiments using both simulated and real world data. The results show that the optimization of a utility estimate such as the cross-validation (CV) score is liable to finding overfitted models due to relatively high variance in the utility estimates when the data is scarce. This can also lead to substantial selection induced bias and optimism in the performance evaluation for the selected model. From a predictive viewpoint, best results are obtained by accounting for model uncertainty by forming the full encompassing model, such as the Bayesian model averaging solution over the candidate models. If the encompassing model is too complex, it can be robustly simplified by the projection method, in which the information of the full model is projected onto the submodels. This approach is substantially less prone to overfitting than selection based on CV-score. Overall, the projection method appears to outperform also the maximum a posteriori model and the selection of the most probable variables. The study also demonstrates that the model selection can greatly benefit from using cross-validation outside the searching process both for guiding the model size selection and assessing the predictive performance of the finally selected model.</subfield></datafield><datafield tag="650" ind1=" " ind2="4"><subfield code="a">Bayesian model selection</subfield></datafield><datafield tag="650" ind1=" " ind2="4"><subfield code="a">Cross-validation</subfield></datafield><datafield tag="650" ind1=" " ind2="4"><subfield code="a">Reference model</subfield></datafield><datafield tag="650" ind1=" " ind2="4"><subfield code="a">Projection</subfield></datafield><datafield tag="650" ind1=" " ind2="4"><subfield code="a">Selection bias</subfield></datafield><datafield tag="700" ind1="1" ind2=" "><subfield code="a">Vehtari, Aki</subfield><subfield code="4">aut</subfield></datafield><datafield tag="773" ind1="0" ind2="8"><subfield code="i">Enthalten in</subfield><subfield code="t">Statistics and computing</subfield><subfield code="d">Springer US, 1991</subfield><subfield code="g">27(2016), 3 vom: 07. Apr., Seite 711-735</subfield><subfield code="w">(DE-627)131007963</subfield><subfield code="w">(DE-600)1087487-2</subfield><subfield code="w">(DE-576)052732894</subfield><subfield code="x">0960-3174</subfield><subfield code="7">nnns</subfield></datafield><datafield tag="773" ind1="1" ind2="8"><subfield code="g">volume:27</subfield><subfield code="g">year:2016</subfield><subfield code="g">number:3</subfield><subfield code="g">day:07</subfield><subfield code="g">month:04</subfield><subfield code="g">pages:711-735</subfield></datafield><datafield tag="856" ind1="4" ind2="1"><subfield code="u">https://doi.org/10.1007/s11222-016-9649-y</subfield><subfield code="z">lizenzpflichtig</subfield><subfield code="3">Volltext</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_USEFLAG_A</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">SYSFLAG_A</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_OLC</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">SSG-OLC-TEC</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">SSG-OLC-MAT</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_ILN_70</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_ILN_4126</subfield></datafield><datafield tag="951" ind1=" " ind2=" "><subfield code="a">AR</subfield></datafield><datafield tag="952" ind1=" " ind2=" "><subfield code="d">27</subfield><subfield code="j">2016</subfield><subfield code="e">3</subfield><subfield code="b">07</subfield><subfield code="c">04</subfield><subfield code="h">711-735</subfield></datafield></record></collection>
|
score |
7.4005537 |