Speakers - 2026

Title: Supervised machine learning for predicting low risk dengue: A clinical data driven approach for appropriate and value based healthcare

Abstract

Background: Dengue fever represents the majority of dengue virus infections and is typically characterized by self-limiting febrile illness without progression to severe complications. Despite its benign course, accurate risk stratification remains critical to prevent unnecessary hospitalisations and to optimise resource allocation, especially in endemic regions. Existing clinical criteria help guide management decisions, but their utility varies across different healthcare settings and populations. This study aims to identify low-risk dengue patients suitable for outpatient management to reduce healthcare burden and support appropriate care placement.

Methods: Clinical and laboratory data comprising 35 features were collected from 530 dengue patients in Sri Lanka, classified per WHO criteria into 407 low-risk and 123 high-risk cases. Variables included age, sex, comorbidities, hypoalbuminaemia, haematocrit increase, and platelet count. Multiple machine learning models—including Logistic Regression, SVM, KNN, Random Forest, and XGBoost—were trained to predict low-risk patients. Model performance was assessed using AUC, F1 score, recall, precision.

Results: Following hyperparameter tuning and feature selection, 22 features1 (12 categorical and 10 continuous) were selected for model development. Among the evaluated models, the CatBoostClassifier achieved the best overall performance, with an AUC of 0.88 and an F1-score of 0.89, supported by balanced precision and recall values (both 0.89). The AdaBoostClassifier also demonstrated competitive performance, particularly with a strong F1-score (0.9) and a relatively high AUC 0.87, highlighting its effectiveness as an alternative approach.

Conclusion: Our findings highlight the potential of machine learning models —particularly the CatBoost algorithm in predicting low-risk dengue cases. By accurately identifying patients safe for outpatient management, this model can significantly optimize healthcare resource allocation and reduce the overall burden on the healthcare system. The next phases of this research will focus on external validation of our models using an independent patient dataset and prospective validation within a proposed clinical workflow.

What will the audience take away from presentation:

1. A practical understanding of how machine learning models can be applied to clinical decision-making in infectious diseases.
2. Insights into a predictive tool that improves dengue triage accuracy, ensuring safer outpatient management.
3. Evidence on how data-driven tools can reduce unnecessary hospitalizations, saving healthcare resources.
4. Knowledge of how this approach can be adapted to other conditions beyond dengue, broadening its utility.
5. A framework for integrating predictive analytics into routine hospital protocols without disrupting clinical workflows.