Early Sepsis Prediction in ICU Patients using a Transformer-XGBoost Hybrid Model

View at publisher Summarise with VIRAT
Authors: J. Gnanasoundharam, ANANDIKAMATHI R, ASEEMA SAFNA
Cited by: 0
References: 18

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Sepsis is a critical medical condition that continues to be a substantial contributor to mortality in intensive care units (ICUs) primarily due to the time lost in accessing timely detection and treatment. Recent advances in artificial intelligence,mainly in deep learning and ensemble models, have been employed to model complex physiological data. Transformer-based models have been shown to be useful in modeling and capturing long-range temporal dependencies in data streams, while gradient boosting models like XGBoost have shown to provide robust performance rates when developing structured clinical features. Explainable AI (XAI) models presents a critical development in addressing concerns about models being interpretable for various stakeholders involved with the clinical therapy-related decision-making process. In this work, a hybrid framework for early sepsis prediction is proposed that integrates a transformer-based time-series encoder within an XGBoost classifier. The time-series encoder captures temporal patterns present in continuous ICU monitoring dataset and inputs the prediction data into XGBoost with structured clinical variables to improve performance. The SHAP based explainer is further utilized to highlight the top features resulting in prediction of risk. Experiments conducted on the MIMIC-III dataset demonstrate that the proposed hybrid model outperforms standard benchmark approaches for early sepsis prediction. The hybrid time-series sequence model achieves higher ROC-AUC and PR-AUC measures compared to each of the individual benchmark data-driven models evaluated in our study.

Fields of Study & Subjects

Machine Learning in Healthcare Sepsis Diagnosis and Treatment Artificial Intelligence in Healthcare

Publication details

Journal / source
International Conference on Communication, Computing and Electronics Systems (ICCCES)
Pages
1521–1526
Publication date
21 January 2026
Language
EN