{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,9,24]],"date-time":"2026-09-24T06:02:56Z","timestamp":1790229776993,"version":"4.1.0"},"reference-count":33,"publisher":"Frontiers Media SA","license":[{"start":{"date-parts":[[2026,9,24]],"date-time":"2026-09-24T00:00:00Z","timestamp":1790208000000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"content-domain":{"domain":["frontiersin.org"],"crossmark-restriction":true},"short-container-title":["Front. Artif. Intell."],"abstract":"<jats:sec>\n                    <jats:title>Introduction<\/jats:title>\n                    <jats:p>Accurate forecasting of physiological time-series in the intensive care unit (ICU) is critical for early-warning systems, deterioration detection, and dynamic resource allocation. Yet the comparative literature suffers from five persistent methodological deficiencies: non-unified experimental conditions, heterogeneous preprocessing, univariate classical baselines compared against multivariate neural models, single-run point estimates without variance, and underspecified prediction tasks.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Methods<\/jats:title>\n                    <jats:p>\n                      This paper addresses all five deficiencies through a standardized, reproducible benchmarking study comparing Autoregressive Integrated Moving Average (ARIMA), Vector Autoregression (VAR), Artificial Neural Networks (ANN), Gated Recurrent Units (GRU), and Long Short-Term Memory (LSTM) networks on one-step-ahead (1-h-ahead) prediction of heart rate, respiratory rate, systolic and diastolic blood pressure, and oxygen saturation across two large-scale ICU cohorts (8,000 MIMIC-III and 15,000 eICU admissions). VAR is introduced as a multivariate linear baseline alongside univariate ARIMA, enabling a three-way decomposition of performance gains into multivariate modeling, nonlinear approximation, and gated temporal memory. All experiments are repeated over 10 independent seeds and reported as mean \u00b1 95% confidence interval with Diebold-Mariano, paired\n                      <jats:italic>t<\/jats:italic>\n                      -test, and Bonferroni-corrected ANOVA significance testing. Seven categories of ablation study, a clinical alarm simulation, a leave-one-hospital-out protocol, and a subgroup fairness analysis are additionally conducted.\n                    <\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Results<\/jats:title>\n                    <jats:p>\n                      LSTM achieves the highest performance on MIMIC-III (RMSE 0.082 \u00b1 0.004,\n                      <jats:italic>R<\/jats:italic>\n                      <jats:sup>2<\/jats:sup>\n                      0.940 \u00b1 0.008, directional accuracy 90.1%\u00b10.9%), with GRU reaching near-equivalent accuracy at 64% of training cost. The layered decomposition reveals that gated temporal memory provides the largest marginal gain (\u221247% to \u221253% RMSE over ANN), followed by nonlinear approximation (\u221219% over VAR) and multivariate modeling (\u221213% over ARIMA). Ablation identifies normalization as the single most critical preprocessing step (+79.3% RMSE without it) and a window length of\n                      <jats:italic>l<\/jats:italic>\n                      = 10 as optimal. The clinical alarm simulation demonstrates a 38.4% false alarm rate reduction for LSTM relative to threshold-only monitoring, and the leave-one-hospital-out experiment confirms cross-institutional generalizability with consistent model ranking.\n                    <\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Discussion<\/jats:title>\n                    <jats:p>Subgroup fairness analysis identifies a performance gap for severely ill patients (Sequential Organ Failure Assessment score \u2265 8, ratio 1.27) warranting targeted future work. These results establish a reusable, methodologically rigorous benchmark for ICU physiological forecasting and provide actionable model-selection guidance for clinical deployment.<\/jats:p>\n                  <\/jats:sec>","DOI":"10.3389\/frai.2026.1873285","type":"journal-article","created":{"date-parts":[[2026,9,24]],"date-time":"2026-09-24T05:44:51Z","timestamp":1790228691000},"update-policy":"https:\/\/doi.org\/10.3389\/crossmark-policy","source":"Crossref","is-referenced-by-count":0,"title":["Performance evaluation of ARIMA, VAR, ANN, GRU, and LSTM for multivariate healthcare time-series prediction: a rigorous, reproducible benchmarking study with ablation analysis on MIMIC-III and eICU"],"prefix":"10.3389","volume":"9","author":[{"given":"Md. Tazul","family":"Islam","sequence":"first","affiliation":[{"name":"Department of Applied Mathematics, Noakhali Science and Technology University","place":["Noakhali, Choumohoni, Bangladesh"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Muhammad","family":"Hanif","sequence":"additional","affiliation":[{"name":"Department of Applied Mathematics, Noakhali Science and Technology University","place":["Noakhali, Choumohoni, Bangladesh"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Samuel","family":"Kim","sequence":"additional","affiliation":[{"name":"Department of Computer Science, Stanford University","place":["Stanford, CA, United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"1965","published-online":{"date-parts":[[2026,9,24]]},"reference":[{"key":"B1","doi-asserted-by":"publisher","first-page":"e21347","DOI":"10.2196\/21347","article-title":"Predicting intensive care unit length of stay and mortality using patient vital signs: machine learning model development and validation","volume":"10","author":"Alghatani","year":"2022","journal-title":"JMIR Med. Inform"},{"key":"B2","doi-asserted-by":"publisher","first-page":"501","DOI":"10.4236\/jamp.2025.132035","article-title":"Time series forecasting in healthcare: comparative study of statistical models and neural networks","volume":"13","author":"Alsheheri","year":"2025","journal-title":"J. Appl. Math. Phys"},{"key":"B3","doi-asserted-by":"publisher","first-page":"279","DOI":"10.1016\/j.jcrc.2020.08.024","article-title":"Early mortality in critical illness-a descriptive analysis of patients who died within 24 hours of ICU admission","volume":"60","author":"Andersen","year":"2020","journal-title":"J. Crit. Care"},{"key":"B4","first-page":"650","article-title":"\u201cA collaborative framework for disease prediction using machine learning,\u201d","volume-title":"Proceedings of the 2025 IEEE international conference on artificial intelligence in engineering and technology (IICAIET)","author":"Baker","year":""},{"key":"B5","doi-asserted-by":"publisher","first-page":"562","DOI":"10.3390\/ijerph22040562","article-title":"Predictive modeling for pandemic forecasting: a COVID-19 study in New Zealand and partner countries","volume":"22","author":"Baker","year":"","journal-title":"Int. J. Environ. Res. Public Health"},{"key":"B6","doi-asserted-by":"publisher","first-page":"109262","DOI":"10.1016\/j.asoc.2022.109181","article-title":"COVID-19 ICU demand forecasting: a two-stage prophet-LSTM hybrid approach","volume":"126","author":"Borges","year":"2022","journal-title":"Appl. Soft Comput"},{"key":"B7","volume-title":"Time Series Analysis: Forecasting and Control","author":"Box","year":"1970"},{"key":"B8","unstructured":"Learning phrase representations using RNN encoder-decoder for statistical machine translation\n          \n          \n            \n              Cho\n              K.\n            \n            \n              van Merrienboer\n              B.\n            \n            \n              Gulcehre\n              C.\n            \n            \n              Bahdanau\n              D.\n            \n            \n              Bougares\n              F.\n            \n            \n              Schwenk\n              H.\n            \n          \n          arXiv\n          \n          2014"},{"key":"B9","doi-asserted-by":"publisher","first-page":"15","DOI":"10.1111\/j.1469-8137.2007.02055.x","article-title":"Forecasting health care: predicting health outcomes with excellent accuracy using ANN","volume":"52","author":"Dickie","year":"2007","journal-title":"J. Healthc. Manage"},{"key":"B10","doi-asserted-by":"publisher","first-page":"253","DOI":"10.1080\/07350015.1995.10524599","article-title":"Comparing predictive accuracy","volume":"13","author":"Diebold","year":"1995","journal-title":"J. Bus. Econ. Stat"},{"key":"B11","doi-asserted-by":"publisher","first-page":"104107","DOI":"10.1016\/j.jbi.2022.104107","article-title":"Tell me something interesting: clinical utility of machine learning prediction models in the ICU","volume":"132","author":"Eini-Porat","year":"2022","journal-title":"J. Biomed. Inform"},{"key":"B12","doi-asserted-by":"publisher","first-page":"1","DOI":"10.1109\/CAI54212.2023.00142","article-title":"Prediction of length-of-stay at intensive care unit (ICU) using machine learning based on MIMIC-III database","volume":"12","author":"Hasan","year":"2024","journal-title":"IEEE Access"},{"key":"B13","doi-asserted-by":"publisher","first-page":"1735","DOI":"10.1162\/neco.1997.9.8.1735","article-title":"Long short-term memory","volume":"9","author":"Hochreiter","year":"1997","journal-title":"Neural Comput"},{"key":"B14","doi-asserted-by":"publisher","first-page":"7761","DOI":"10.1111\/nph.17594","article-title":"Prediction of clinical outcomes for ICU patients by stacked ensemble machine learning","volume":"11","author":"Iwase","year":"2021","journal-title":"Eng. Technol. Appl. Sci. Res"},{"key":"B15","doi-asserted-by":"publisher","first-page":"e71247","DOI":"10.2196\/71247","article-title":"A real-time signal-based wavelet long short-term memory method for length-of-stay prediction for the intensive care unit: development and evaluation study","volume":"10","author":"Jiang","year":"2022","journal-title":"JMIR Med. Inform"},{"key":"B16","doi-asserted-by":"publisher","first-page":"160035","DOI":"10.1038\/sdata.2016.35","article-title":"MIMIC-III, a freely accessible critical care database","volume":"3","author":"Johnson","year":"2016","journal-title":"Sci. Data"},{"key":"B17","doi-asserted-by":"publisher","first-page":"S34","DOI":"10.1038\/s41597-023-02041-1","article-title":"Forecasting algorithms in the ICU: challenges and pathways for clinical implementation","volume":"80","author":"Jones","year":"2023","journal-title":"J. Electrocardiol"},{"key":"B18","unstructured":"Adam: a method for stochastic optimization\n          \n          \n            \n              Kingma\n              D. P.\n            \n            \n              Ba\n              J.\n            \n          \n          arXiv\n          \n          2014"},{"key":"B19","doi-asserted-by":"publisher","first-page":"2911","DOI":"10.3828\/idpr.2024.4","article-title":"Enhancing mortality prediction in ICU: a deep learning framework for multivariate time series classification","volume":"28","author":"Kumar","year":"2024","journal-title":"IEEE J. Biomed. Health Inform"},{"key":"B20","doi-asserted-by":"publisher","first-page":"112352","DOI":"10.1029\/2022MS003259","article-title":"A comparative study of ARIMA and LSTM in forecasting time series data","volume":"10","author":"Kumar","year":"2022","journal-title":"IEEE Access"},{"key":"B21","doi-asserted-by":"publisher","first-page":"34512","DOI":"10.1016\/j.scitotenv.2024.175479","article-title":"Time series hybrid prediction model based on deep learning ARIMA-LSTM","volume":"12","author":"Li","year":"2024","journal-title":"IEEE Access"},{"key":"B22","doi-asserted-by":"publisher","first-page":"1748","DOI":"10.1016\/j.ijforecast.2021.03.012","article-title":"Temporal fusion transformers for interpretable multi-horizon time series forecasting","volume":"37","author":"Lim","year":"2021","journal-title":"Int. J. Forecast"},{"key":"B23","doi-asserted-by":"publisher","first-page":"103112","DOI":"10.1016\/j.eclinm.2025.103112","article-title":"Multicenter validation of a machine learning model to predict ICU readmission within 48 hours after discharge","volume":"72","author":"Lim","year":"2025","journal-title":"eClinicalMedicine"},{"key":"B24","first-page":"1273","article-title":"\u201cCommunication-efficient learning of deep networks from decentralized data,\u201d","volume-title":"Proceedings of the 20th International Conference on Artificial Intelligence and Statistics (AISTATS)","author":"McMahan","year":"2017"},{"key":"B25","doi-asserted-by":"publisher","first-page":"180178","DOI":"10.1038\/sdata.2018.178","article-title":"The eICU collaborative research database, a freely available multi-center database for critical care research","volume":"5","author":"Pollard","year":"2018","journal-title":"Sci. Data"},{"key":"B26","doi-asserted-by":"publisher","first-page":"1201","DOI":"10.1038\/s41579-023-00900-7","article-title":"Neural networks for healthcare time-series: performance and challenges","volume":"27","author":"Singh","year":"2023","journal-title":"IEEE J. Biomed. Health Inform"},{"key":"B27","first-page":"1394","article-title":"\u201cA comparison of ARIMA and LSTM in forecasting time series,\u201d","volume-title":"2018 17th IEEE international conference on machine learning and applications (ICMLA)","author":"Wang","year":"2019"},{"key":"B28","doi-asserted-by":"publisher","first-page":"1","DOI":"10.1016\/j.artmed.2018.08.004","article-title":"Predicting ICU readmission using grouped physiological and medication trends","volume":"93","author":"Xue","year":"2018","journal-title":"Artif. Intell. Med"},{"key":"B29","doi-asserted-by":"publisher","first-page":"104969","DOI":"10.1016\/j.jbi.2025.104969","article-title":"Graph-SPA: a spatiotemporal graph neural network based framework for ARDS prediction and interpretability","volume":"162","author":"Yadav","year":"2025","journal-title":"J. Biomed. Inform"},{"key":"B30","first-page":"1","article-title":"\u201cForecastICU: a prognostic decision support system for ICU patients,\u201d","volume-title":"Proceedings of the 33rd international conference on machine learning","author":"Yoon","year":"2016"},{"key":"B31","first-page":"1","article-title":"\u201cMonitoring ICU mortality risk with a long short-term memory deep learning approach,\u201d","volume-title":"AMIA annual symposium proceedings","author":"Yu","year":"2019"},{"key":"B32","doi-asserted-by":"publisher","first-page":"S28","DOI":"10.1186\/s12991-023-00462-1","article-title":"Forecasting ICU census using time-series and survival models","volume":"80","author":"Zhao","year":"2023","journal-title":"J. Electrocardiol"},{"key":"B33","article-title":"\u201cTime series forecasting based on ARIMA and LSTM models,\u201d","volume-title":"Proceedings of the IEEE international conference on systems, man, and cybernetics, Kuching, Sarawak, Malaysia (6\u201310 October, 2024)","author":"Zhou","year":"2024"}],"container-title":["Frontiers in Artificial Intelligence"],"original-title":[],"link":[{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/frai.2026.1873285\/full","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,9,24]],"date-time":"2026-09-24T05:44:54Z","timestamp":1790228694000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/frai.2026.1873285\/full"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,9,24]]},"references-count":33,"alternative-id":["10.3389\/frai.2026.1873285"],"URL":"https:\/\/doi.org\/10.3389\/frai.2026.1873285","relation":{},"ISSN":["2624-8212"],"issn-type":[{"value":"2624-8212","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,9,24]]},"article-number":"1873285"}}