Balancing Accuracy and Interpretability in Machine Learning–Based Early Detection of Patient Deterioration
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
Patient deterioration is an issue that requires early detection at hospital wards where nurses constantly observe the patient vital signs. This paper presents a predictive algorithmic model of patient deterioration based on physiological parameters collected regularly using machine learning. A clinically realistic synthetic dataset was generated through clinically accepted ranges of heart rate, respiratory rate, blood pressure, body temperature and oxygen saturation. The data has been normalized and separated into training and test data using 70:30 split. The support vector machine (SVM), k-nearest neighbors (KNN), and Decision Tree are all classifiers that were implemented and compared. Accuracy, confusion matrix analysis, receiver operating characteristic (ROC) curves and area under the curve (AUC) were used to measure model performance. The cross-validation was conducted five times to determine the model robustness and feature importance analysis was done to facilitate clinical interpretability. Experimental results indicate that the Decision Tree classifier achieves the highest classification accuracy, while the Support Vector Machine demonstrates strong discriminative capability with a lower false-negative rate. Although the Decision Tree model provides superior predictive performance, SVM offers better robustness and generalization. The proposed framework balances predictive accuracy and clinical interpretability, making it suitable for integration into nursing decision-support systems.
Fields of Study & Subjects
Artificial Intelligence in Healthcare
Healthcare Technology and Patient Monitoring
Machine Learning in Healthcare
Publication details
- Journal / source
- International Conference on Sentiment Analysis and Deep Learning (ICSADL)
- Pages
- 1103–1108
- Publication date
- 18 February 2026
- Language
- EN