WindowSHAP: An efficient framework for explaining time-series classifiers based on Shapley values
[Display omitted] Unpacking and comprehending how black-box machine learning algorithms (such as deep learning models) make decisions has been a persistent challenge for researchers and end-users. Explaining time-series predictive models is useful for clinical applications with high stakes to unders...
Gespeichert in:
Veröffentlicht in: | Journal of biomedical informatics 2023-08, Vol.144, p.104438-104438, Article 104438 |
---|---|
Hauptverfasser: | , , , , |
Format: | Artikel |
Sprache: | eng |
Schlagworte: | |
Online-Zugang: | Volltext |
Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
container_end_page | 104438 |
---|---|
container_issue | |
container_start_page | 104438 |
container_title | Journal of biomedical informatics |
container_volume | 144 |
creator | Nayebi, Amin Tipirneni, Sindhu Reddy, Chandan K. Foreman, Brandon Subbian, Vignesh |
description | [Display omitted]
Unpacking and comprehending how black-box machine learning algorithms (such as deep learning models) make decisions has been a persistent challenge for researchers and end-users. Explaining time-series predictive models is useful for clinical applications with high stakes to understand the behavior of prediction models, e.g., to determine how different variables and time points influence the clinical outcome. However, existing approaches to explain such models are frequently unique to architectures and data where the features do not have a time-varying component. In this paper, we introduce WindowSHAP, a model-agnostic framework for explaining time-series classifiers using Shapley values. We intend for WindowSHAP to mitigate the computational complexity of calculating Shapley values for long time-series data as well as improve the quality of explanations. WindowSHAP is based on partitioning a sequence into time windows. Under this framework, we present three distinct algorithms of Stationary, Sliding and Dynamic WindowSHAP, each evaluated against baseline approaches, KernelSHAP and TimeSHAP, using perturbation and sequence analyses metrics. We applied our framework to clinical time-series data from both a specialized clinical domain (Traumatic Brain Injury - TBI) as well as a broad clinical domain (critical care medicine). The experimental results demonstrate that, based on the two quantitative metrics, our framework is superior at explaining clinical time-series classifiers, while also reducing the complexity of computations. We show that for time-series data with 120 time steps (hours), merging 10 adjacent time points can reduce the CPU time of WindowSHAP by 80 % compared to KernelSHAP. We also show that our Dynamic WindowSHAP algorithm focuses more on the most important time steps and provides more understandable explanations. As a result, WindowSHAP not only accelerates the calculation of Shapley values for time-series data, but also delivers more understandable explanations with higher quality. |
doi_str_mv | 10.1016/j.jbi.2023.104438 |
format | Article |
fullrecord | <record><control><sourceid>proquest_pubme</sourceid><recordid>TN_cdi_pubmedcentral_primary_oai_pubmedcentral_nih_gov_10552726</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><els_id>S1532046423001594</els_id><sourcerecordid>2835275326</sourcerecordid><originalsourceid>FETCH-LOGICAL-c452t-b95d1acb3417f662c42e3a08356d99f450ff8ba8a75cf52fba304cbb669632423</originalsourceid><addsrcrecordid>eNp9kc1uEzEUhS0EoqXwAGyQl2wm-H8msEBRVShSJZAKYmlde65bh5lxsCcpfXscpUSwYWVbPvf4-HyEvORswRk3b9aLtYsLwYSsZ6Vk94icci1Fw1THHh_3Rp2QZ6WsGeNca_OUnMhWcSVNd0rge5z6dHd9ufrylq4miiFEH3Gaacgw4l3KP2hImeKvzQBxitMNneOITcEcsVA_QCkxRMyFOijY0zTR61vYDHhPdzBssTwnTwIMBV88rGfk24eLr-eXzdXnj5_OV1eNV1rMjVvqnoN3UvE2GCO8EiiBdVKbfrkMSrMQOgcdtNoHLYIDyZR3zpilkUIJeUbeH3w3Wzdi7-sfMgx2k-MI-d4miPbfmyne2pu0s5xpLVphqsPrB4ecftbksx1j8TgMMGHaFitqGNHWTvdSfpD6nErJGI7vcGb3bOzaVjZ2z8Ye2NSZV38HPE78gVEF7w4CrDXtaqe27FF47GNGP9s-xf_Y_wZgcKCt</addsrcrecordid><sourcetype>Open Access Repository</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype><pqid>2835275326</pqid></control><display><type>article</type><title>WindowSHAP: An efficient framework for explaining time-series classifiers based on Shapley values</title><source>MEDLINE</source><source>Access via ScienceDirect (Elsevier)</source><source>EZB-FREE-00999 freely available EZB journals</source><creator>Nayebi, Amin ; Tipirneni, Sindhu ; Reddy, Chandan K. ; Foreman, Brandon ; Subbian, Vignesh</creator><creatorcontrib>Nayebi, Amin ; Tipirneni, Sindhu ; Reddy, Chandan K. ; Foreman, Brandon ; Subbian, Vignesh</creatorcontrib><description>[Display omitted]
Unpacking and comprehending how black-box machine learning algorithms (such as deep learning models) make decisions has been a persistent challenge for researchers and end-users. Explaining time-series predictive models is useful for clinical applications with high stakes to understand the behavior of prediction models, e.g., to determine how different variables and time points influence the clinical outcome. However, existing approaches to explain such models are frequently unique to architectures and data where the features do not have a time-varying component. In this paper, we introduce WindowSHAP, a model-agnostic framework for explaining time-series classifiers using Shapley values. We intend for WindowSHAP to mitigate the computational complexity of calculating Shapley values for long time-series data as well as improve the quality of explanations. WindowSHAP is based on partitioning a sequence into time windows. Under this framework, we present three distinct algorithms of Stationary, Sliding and Dynamic WindowSHAP, each evaluated against baseline approaches, KernelSHAP and TimeSHAP, using perturbation and sequence analyses metrics. We applied our framework to clinical time-series data from both a specialized clinical domain (Traumatic Brain Injury - TBI) as well as a broad clinical domain (critical care medicine). The experimental results demonstrate that, based on the two quantitative metrics, our framework is superior at explaining clinical time-series classifiers, while also reducing the complexity of computations. We show that for time-series data with 120 time steps (hours), merging 10 adjacent time points can reduce the CPU time of WindowSHAP by 80 % compared to KernelSHAP. We also show that our Dynamic WindowSHAP algorithm focuses more on the most important time steps and provides more understandable explanations. As a result, WindowSHAP not only accelerates the calculation of Shapley values for time-series data, but also delivers more understandable explanations with higher quality.</description><identifier>ISSN: 1532-0464</identifier><identifier>ISSN: 1532-0480</identifier><identifier>EISSN: 1532-0480</identifier><identifier>DOI: 10.1016/j.jbi.2023.104438</identifier><identifier>PMID: 37414368</identifier><language>eng</language><publisher>United States: Elsevier Inc</publisher><subject>Algorithms ; Benchmarking ; Brain Injuries, Traumatic - diagnosis ; Explainable artificial intelligence ; Humans ; Machine Learning ; Model interpretation ; Shapley value ; Time Factors ; Time-series data</subject><ispartof>Journal of biomedical informatics, 2023-08, Vol.144, p.104438-104438, Article 104438</ispartof><rights>2023 Elsevier Inc.</rights><rights>Copyright © 2023 Elsevier Inc. All rights reserved.</rights><lds50>peer_reviewed</lds50><oa>free_for_read</oa><woscitedreferencessubscribed>false</woscitedreferencessubscribed><citedby>FETCH-LOGICAL-c452t-b95d1acb3417f662c42e3a08356d99f450ff8ba8a75cf52fba304cbb669632423</citedby><cites>FETCH-LOGICAL-c452t-b95d1acb3417f662c42e3a08356d99f450ff8ba8a75cf52fba304cbb669632423</cites><orcidid>0000-0002-5418-674X ; 0000-0001-9974-8382 ; 0000-0001-5502-1616 ; 0000-0003-2839-3662 ; 0000-0002-0656-6748</orcidid></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><linktohtml>$$Uhttps://dx.doi.org/10.1016/j.jbi.2023.104438$$EHTML$$P50$$Gelsevier$$Hfree_for_read</linktohtml><link.rule.ids>230,314,780,784,885,3550,27924,27925,45995</link.rule.ids><backlink>$$Uhttps://www.ncbi.nlm.nih.gov/pubmed/37414368$$D View this record in MEDLINE/PubMed$$Hfree_for_read</backlink></links><search><creatorcontrib>Nayebi, Amin</creatorcontrib><creatorcontrib>Tipirneni, Sindhu</creatorcontrib><creatorcontrib>Reddy, Chandan K.</creatorcontrib><creatorcontrib>Foreman, Brandon</creatorcontrib><creatorcontrib>Subbian, Vignesh</creatorcontrib><title>WindowSHAP: An efficient framework for explaining time-series classifiers based on Shapley values</title><title>Journal of biomedical informatics</title><addtitle>J Biomed Inform</addtitle><description>[Display omitted]
Unpacking and comprehending how black-box machine learning algorithms (such as deep learning models) make decisions has been a persistent challenge for researchers and end-users. Explaining time-series predictive models is useful for clinical applications with high stakes to understand the behavior of prediction models, e.g., to determine how different variables and time points influence the clinical outcome. However, existing approaches to explain such models are frequently unique to architectures and data where the features do not have a time-varying component. In this paper, we introduce WindowSHAP, a model-agnostic framework for explaining time-series classifiers using Shapley values. We intend for WindowSHAP to mitigate the computational complexity of calculating Shapley values for long time-series data as well as improve the quality of explanations. WindowSHAP is based on partitioning a sequence into time windows. Under this framework, we present three distinct algorithms of Stationary, Sliding and Dynamic WindowSHAP, each evaluated against baseline approaches, KernelSHAP and TimeSHAP, using perturbation and sequence analyses metrics. We applied our framework to clinical time-series data from both a specialized clinical domain (Traumatic Brain Injury - TBI) as well as a broad clinical domain (critical care medicine). The experimental results demonstrate that, based on the two quantitative metrics, our framework is superior at explaining clinical time-series classifiers, while also reducing the complexity of computations. We show that for time-series data with 120 time steps (hours), merging 10 adjacent time points can reduce the CPU time of WindowSHAP by 80 % compared to KernelSHAP. We also show that our Dynamic WindowSHAP algorithm focuses more on the most important time steps and provides more understandable explanations. As a result, WindowSHAP not only accelerates the calculation of Shapley values for time-series data, but also delivers more understandable explanations with higher quality.</description><subject>Algorithms</subject><subject>Benchmarking</subject><subject>Brain Injuries, Traumatic - diagnosis</subject><subject>Explainable artificial intelligence</subject><subject>Humans</subject><subject>Machine Learning</subject><subject>Model interpretation</subject><subject>Shapley value</subject><subject>Time Factors</subject><subject>Time-series data</subject><issn>1532-0464</issn><issn>1532-0480</issn><issn>1532-0480</issn><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2023</creationdate><recordtype>article</recordtype><sourceid>EIF</sourceid><recordid>eNp9kc1uEzEUhS0EoqXwAGyQl2wm-H8msEBRVShSJZAKYmlde65bh5lxsCcpfXscpUSwYWVbPvf4-HyEvORswRk3b9aLtYsLwYSsZ6Vk94icci1Fw1THHh_3Rp2QZ6WsGeNca_OUnMhWcSVNd0rge5z6dHd9ufrylq4miiFEH3Gaacgw4l3KP2hImeKvzQBxitMNneOITcEcsVA_QCkxRMyFOijY0zTR61vYDHhPdzBssTwnTwIMBV88rGfk24eLr-eXzdXnj5_OV1eNV1rMjVvqnoN3UvE2GCO8EiiBdVKbfrkMSrMQOgcdtNoHLYIDyZR3zpilkUIJeUbeH3w3Wzdi7-sfMgx2k-MI-d4miPbfmyne2pu0s5xpLVphqsPrB4ecftbksx1j8TgMMGHaFitqGNHWTvdSfpD6nErJGI7vcGb3bOzaVjZ2z8Ye2NSZV38HPE78gVEF7w4CrDXtaqe27FF47GNGP9s-xf_Y_wZgcKCt</recordid><startdate>20230801</startdate><enddate>20230801</enddate><creator>Nayebi, Amin</creator><creator>Tipirneni, Sindhu</creator><creator>Reddy, Chandan K.</creator><creator>Foreman, Brandon</creator><creator>Subbian, Vignesh</creator><general>Elsevier Inc</general><scope>6I.</scope><scope>AAFTH</scope><scope>CGR</scope><scope>CUY</scope><scope>CVF</scope><scope>ECM</scope><scope>EIF</scope><scope>NPM</scope><scope>AAYXX</scope><scope>CITATION</scope><scope>7X8</scope><scope>5PM</scope><orcidid>https://orcid.org/0000-0002-5418-674X</orcidid><orcidid>https://orcid.org/0000-0001-9974-8382</orcidid><orcidid>https://orcid.org/0000-0001-5502-1616</orcidid><orcidid>https://orcid.org/0000-0003-2839-3662</orcidid><orcidid>https://orcid.org/0000-0002-0656-6748</orcidid></search><sort><creationdate>20230801</creationdate><title>WindowSHAP: An efficient framework for explaining time-series classifiers based on Shapley values</title><author>Nayebi, Amin ; Tipirneni, Sindhu ; Reddy, Chandan K. ; Foreman, Brandon ; Subbian, Vignesh</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-c452t-b95d1acb3417f662c42e3a08356d99f450ff8ba8a75cf52fba304cbb669632423</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2023</creationdate><topic>Algorithms</topic><topic>Benchmarking</topic><topic>Brain Injuries, Traumatic - diagnosis</topic><topic>Explainable artificial intelligence</topic><topic>Humans</topic><topic>Machine Learning</topic><topic>Model interpretation</topic><topic>Shapley value</topic><topic>Time Factors</topic><topic>Time-series data</topic><toplevel>peer_reviewed</toplevel><toplevel>online_resources</toplevel><creatorcontrib>Nayebi, Amin</creatorcontrib><creatorcontrib>Tipirneni, Sindhu</creatorcontrib><creatorcontrib>Reddy, Chandan K.</creatorcontrib><creatorcontrib>Foreman, Brandon</creatorcontrib><creatorcontrib>Subbian, Vignesh</creatorcontrib><collection>ScienceDirect Open Access Titles</collection><collection>Elsevier:ScienceDirect:Open Access</collection><collection>Medline</collection><collection>MEDLINE</collection><collection>MEDLINE (Ovid)</collection><collection>MEDLINE</collection><collection>MEDLINE</collection><collection>PubMed</collection><collection>CrossRef</collection><collection>MEDLINE - Academic</collection><collection>PubMed Central (Full Participant titles)</collection><jtitle>Journal of biomedical informatics</jtitle></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext</fulltext></delivery><addata><au>Nayebi, Amin</au><au>Tipirneni, Sindhu</au><au>Reddy, Chandan K.</au><au>Foreman, Brandon</au><au>Subbian, Vignesh</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>WindowSHAP: An efficient framework for explaining time-series classifiers based on Shapley values</atitle><jtitle>Journal of biomedical informatics</jtitle><addtitle>J Biomed Inform</addtitle><date>2023-08-01</date><risdate>2023</risdate><volume>144</volume><spage>104438</spage><epage>104438</epage><pages>104438-104438</pages><artnum>104438</artnum><issn>1532-0464</issn><issn>1532-0480</issn><eissn>1532-0480</eissn><abstract>[Display omitted]
Unpacking and comprehending how black-box machine learning algorithms (such as deep learning models) make decisions has been a persistent challenge for researchers and end-users. Explaining time-series predictive models is useful for clinical applications with high stakes to understand the behavior of prediction models, e.g., to determine how different variables and time points influence the clinical outcome. However, existing approaches to explain such models are frequently unique to architectures and data where the features do not have a time-varying component. In this paper, we introduce WindowSHAP, a model-agnostic framework for explaining time-series classifiers using Shapley values. We intend for WindowSHAP to mitigate the computational complexity of calculating Shapley values for long time-series data as well as improve the quality of explanations. WindowSHAP is based on partitioning a sequence into time windows. Under this framework, we present three distinct algorithms of Stationary, Sliding and Dynamic WindowSHAP, each evaluated against baseline approaches, KernelSHAP and TimeSHAP, using perturbation and sequence analyses metrics. We applied our framework to clinical time-series data from both a specialized clinical domain (Traumatic Brain Injury - TBI) as well as a broad clinical domain (critical care medicine). The experimental results demonstrate that, based on the two quantitative metrics, our framework is superior at explaining clinical time-series classifiers, while also reducing the complexity of computations. We show that for time-series data with 120 time steps (hours), merging 10 adjacent time points can reduce the CPU time of WindowSHAP by 80 % compared to KernelSHAP. We also show that our Dynamic WindowSHAP algorithm focuses more on the most important time steps and provides more understandable explanations. As a result, WindowSHAP not only accelerates the calculation of Shapley values for time-series data, but also delivers more understandable explanations with higher quality.</abstract><cop>United States</cop><pub>Elsevier Inc</pub><pmid>37414368</pmid><doi>10.1016/j.jbi.2023.104438</doi><tpages>1</tpages><orcidid>https://orcid.org/0000-0002-5418-674X</orcidid><orcidid>https://orcid.org/0000-0001-9974-8382</orcidid><orcidid>https://orcid.org/0000-0001-5502-1616</orcidid><orcidid>https://orcid.org/0000-0003-2839-3662</orcidid><orcidid>https://orcid.org/0000-0002-0656-6748</orcidid><oa>free_for_read</oa></addata></record> |
fulltext | fulltext |
identifier | ISSN: 1532-0464 |
ispartof | Journal of biomedical informatics, 2023-08, Vol.144, p.104438-104438, Article 104438 |
issn | 1532-0464 1532-0480 1532-0480 |
language | eng |
recordid | cdi_pubmedcentral_primary_oai_pubmedcentral_nih_gov_10552726 |
source | MEDLINE; Access via ScienceDirect (Elsevier); EZB-FREE-00999 freely available EZB journals |
subjects | Algorithms Benchmarking Brain Injuries, Traumatic - diagnosis Explainable artificial intelligence Humans Machine Learning Model interpretation Shapley value Time Factors Time-series data |
title | WindowSHAP: An efficient framework for explaining time-series classifiers based on Shapley values |
url | https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2024-12-23T02%3A22%3A36IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-proquest_pubme&rft_val_fmt=info:ofi/fmt:kev:mtx:journal&rft.genre=article&rft.atitle=WindowSHAP:%20An%20efficient%20framework%20for%20explaining%20time-series%20classifiers%20based%20on%20Shapley%20values&rft.jtitle=Journal%20of%20biomedical%20informatics&rft.au=Nayebi,%20Amin&rft.date=2023-08-01&rft.volume=144&rft.spage=104438&rft.epage=104438&rft.pages=104438-104438&rft.artnum=104438&rft.issn=1532-0464&rft.eissn=1532-0480&rft_id=info:doi/10.1016/j.jbi.2023.104438&rft_dat=%3Cproquest_pubme%3E2835275326%3C/proquest_pubme%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_pqid=2835275326&rft_id=info:pmid/37414368&rft_els_id=S1532046423001594&rfr_iscdi=true |