Authors :
Manish Raj; Goalla Kartheek; Arikatla Sai Sumedha; Chintala Guru Sunanda; Bhagyashree G Malipatil; Poojitha S; Praveen Kumar Burra; Bharani Kumar Depuru
Volume/Issue :
Volume 10 - 2025, Issue 6 - June
Google Scholar :
https://tinyurl.com/yvryvaf6
DOI :
https://doi.org/10.38124/ijisrt/25jun278
Note : A published paper may take 4-5 working days from the publication date to appear in PlumX Metrics, Semantic Scholar, and ResearchGate.
Abstract :
Real-time monitoring and prediction of physiological parameters in critical care settings remains essential for
preventing patient deterioration and enabling timely medical interventions. This research examines various computational
approaches for short-term prediction of six critical vital signs: Heart Rate (HR), Systolic/Diastolic Blood Pressure
(SBP/DBP), Respiratory Rate (RR), Oxygen Saturation (SpO2), and Temperature (Temp) using the VitalDB clinical
database. Our investigation began with conventional time-series methods including Autoregressive Integrated Moving
Average (ARIMA) models and progressed to sophisticated neural architectures such as Long Short-Term Memory (LSTM)
networks. However, these approaches demonstrated limitations in modeling complex relationships between multiple
physiological variables simultaneously. Subsequently, we implemented advanced hybrid architectures incorporating
Bidirectional LSTM (BiLSTM) layers, Convolutional Neural Networks (CNN), and Graph Attention Network (GAT)
mechanisms. Although this hybrid model achieved enhanced prediction accuracy, their computational complexity posed
challenges for clinical deployment. Addressing practical implementation requirements, we evaluated Multi-Layer
Perceptron (MLP)-based frameworks, specifically Patch Time Series Transformer (PatchTST) and Time Series Mixer
(TSMixer) architectures. PatchTST effectively captures extended temporal dependencies but lacks comprehensive cross-
variable interaction modeling. Conversely, TSMixer employs dual mixing mechanisms—temporal and feature-based—to
simultaneously learn chronological patterns and inter-vital correlations. Utilizing 10-minute historical windows to forecast
subsequent 3-minute intervals, TSMixer demonstrated superior performance across all evaluation metrics. The model
achieved the lowest Root Mean Square Error (RMSE) values for all vital parameters while maintaining computational
efficiency suitable for real-time applications. These findings establish TSMixer's potential as a practical solution for
prospective integration into Intensive Care Unit (ICU) monitoring systems, offering both predictive accuracy and
operational feasibility for clinical environments.
Keywords :
Intensive Care Unit, Vital Signs Forecasting, ARIMA, LSTM, BiLSTM, CNN, GAT, PatchTST, TSMixer, Transformer Models, Multivariate Forecasting, Real-Time Health Monitoring, Deep Learning, Clinical Decision Support.
References :
- Jilin Zhang, Lishi Ye, Yongzeng Lai. 2023. Stock Price Prediction Using CNN-BiLSTM-Attention Model. https://www.mdpi.com/2227-7390/11/9/1985.
- Oliver Chojnowski, Dario Luipers, Caterina Neef, Anja Richert. 2023. Forecasting Vital Signs in Human–Robot Collaboration Using Sequence-to-Sequence Models with Bidirectional LSTM: A Comparative Analysis of Uni- and Multi-Variate Approaches. https://www.mdpi.com/2673-4591/58/1/103
- Michael R. Pinsky, Armando Bedoya, Azra Bihorac, Leo Celi, Matthew Churpek, Nicoleta J. Economou-Zavlanos, Paul Elbers, Suchi Saria, Vincent Liu, Patrick G.Lyons, Benjamin Shickel, Patrick Toral, David Tscholl, Gilles Clermont. 2024. Use of artificial intelligence in critical care: opportunities and obstacles. https://ccforum.biomedcentral.com/articles/10.1186/s13054-024-04860-z
- Aswin Thacharodi, Prabhakar Singh, Ramu Meenatchi, Z H Tawfeeq Ahmed, Rejith R S Kumar, Neha V, Sanjana Kavish, Mohsin Maqbool, Saqib Hassan. 2024. Revolutionizing healthcare and medicine: The impact of modern technologies for a healthier future—A comprehensive review. https://pmc.ncbi.nlm.nih.gov/articles/PMC11520245/
- Ugochukwu Orji, Cicek Guven, Dan Stowell. 2025. Enhanced Load Forecasting with GAT-LSTM: Leveraging Grid and Temporal Features. https://arxiv.org/abs/2502.08376
- Tianlong Wu, Feng Chen, Yun Wan. 2018. Graph Attention LSTM Network: A New Model for Traffic Flow Forecasting. https://ieeexplore.ieee.org/document/8612556
- Hussein Ahmad Ahmad, Seyyed Kasra Mortazavi, Mohamed El Bahnasawi, Fadi Al Machot, Witesyavwirwa Vianney Kambale, Kyandoghere Kyamakya. 2024. Enhanced Time Series Forecasting: Integrating PatchTST with BERT Layers. https://ieeexplore.ieee.org/document/10771354
- Si-An Chen, Chun-Liang Li, Nate Yoder, Sercan O. Arik, Tomas Pfister. 2023. TSMixer: An All-MLP Architecture for Time Series Forecasting. https://arxiv.org/abs/2303.06053
- Vijay Ekambaram, Arindam Jati, Nam Nguyen, Phanwadee Sinthong, Jayant Kalagnanam. 2023. TSMixer: Lightweight MLP-Mixer Model for Multivariate Time Series Forecasting. https://arxiv.org/abs/2306.09364
- Haowei Ni, Shuchen Meng, Xieming Geng, Panfeng Li, Zhuoying Li, Xupeng Chen, Xiaotong Wang, Shiyao Zhang. 2024. Time Series Modeling for Heart Rate Prediction: From ARIMA to Transformers. https://arxiv.org/abs/2406.12199
Real-time monitoring and prediction of physiological parameters in critical care settings remains essential for
preventing patient deterioration and enabling timely medical interventions. This research examines various computational
approaches for short-term prediction of six critical vital signs: Heart Rate (HR), Systolic/Diastolic Blood Pressure
(SBP/DBP), Respiratory Rate (RR), Oxygen Saturation (SpO2), and Temperature (Temp) using the VitalDB clinical
database. Our investigation began with conventional time-series methods including Autoregressive Integrated Moving
Average (ARIMA) models and progressed to sophisticated neural architectures such as Long Short-Term Memory (LSTM)
networks. However, these approaches demonstrated limitations in modeling complex relationships between multiple
physiological variables simultaneously. Subsequently, we implemented advanced hybrid architectures incorporating
Bidirectional LSTM (BiLSTM) layers, Convolutional Neural Networks (CNN), and Graph Attention Network (GAT)
mechanisms. Although this hybrid model achieved enhanced prediction accuracy, their computational complexity posed
challenges for clinical deployment. Addressing practical implementation requirements, we evaluated Multi-Layer
Perceptron (MLP)-based frameworks, specifically Patch Time Series Transformer (PatchTST) and Time Series Mixer
(TSMixer) architectures. PatchTST effectively captures extended temporal dependencies but lacks comprehensive cross-
variable interaction modeling. Conversely, TSMixer employs dual mixing mechanisms—temporal and feature-based—to
simultaneously learn chronological patterns and inter-vital correlations. Utilizing 10-minute historical windows to forecast
subsequent 3-minute intervals, TSMixer demonstrated superior performance across all evaluation metrics. The model
achieved the lowest Root Mean Square Error (RMSE) values for all vital parameters while maintaining computational
efficiency suitable for real-time applications. These findings establish TSMixer's potential as a practical solution for
prospective integration into Intensive Care Unit (ICU) monitoring systems, offering both predictive accuracy and
operational feasibility for clinical environments.
Keywords :
Intensive Care Unit, Vital Signs Forecasting, ARIMA, LSTM, BiLSTM, CNN, GAT, PatchTST, TSMixer, Transformer Models, Multivariate Forecasting, Real-Time Health Monitoring, Deep Learning, Clinical Decision Support.
