A Comparative Evaluation Framework for Selecting AI/ML Models for Edge-Enabled Critical-Event Prediction in Intensive Care
Contributors
Narendra Narole
Dr. Lowlesh Nandkishor Yadav
Keywords
Proceeding
Track
General Track
License
Copyright (c) 2026 Sustainable Global Societies Initiative

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
Abstract
Machine learning offers a credible route to earlier detection of clinical deterioration, yet the field has reached an impasse: high predictive performance and real-time edge deployment ability are rarely achieved by the same model. A model that excels on benchmark accuracy may impose high computational complexity, high memory requirements, higher inference latency, poor scalability, and limited suitability for edge devices. This paper presents a research framework that confronts this selection problem directly. Rather than proposing another predictive model, it establishes a common experimental environment in which six representative models such as Logistic Regression, Random Forest, XGBoost, LSTM, GRU, and Transformer are compared under an identical dataset, prediction task, preprocessing strategy, and evaluation protocol. The framework combines predictive-performance analysis across multiple horizons with computational-characteristics assessment and multi-criteria decision analysis to identify the single candidate best suited to Edge Intelligence implementation. We argue that evidence based model selection is a prerequisite not a preliminary for trustworthy, low-latency, alarm-fatigue-resistant clinical decision support.