A Web-Based Machine Learning Application for Classifying Post-COVID Risk among Middle-Aged Adults

Authors

• Himasha Amarathunga, Department of Information Technology, Faculty of Computing, CINEC Campus (Pvt) LtdFollow
• Dhishan Dhammearatchi, Department of Information Technology, Faculty of Computing, CINEC Campus (Pvt) Ltd
• Nimesh Pollwaththage, Department of Information Technology, Faculty of Computing, CINEC Campus (Pvt) Ltd

Document Type

Original Study

Abstract

Post-COVID syndrome (Long-COVID) represents a significant public health challenge, yet accessible tools for individual risk assessment among middle-aged adults remain limited. This study describes the development and evaluation of a web-based machine learning application for the classification of post-COVID risk in adults aged 30–60 years. A publicly available secondary dataset comprising 9,637 patient records was used, of which 5,871 (61%) were classified as high-risk and 3,766 (39%) as standard recovery, reflecting an imbalanced class distribution. The web application incorporates key demographic and clinical features including age, sex, vaccination status and comorbidities to produce a binary risk classification (High/Low). Five machine learning classifiers were evaluated; the optimised Decision Tree Classifier was selected as the final model, achieving an accuracy of 88.8%, sensitivity of 91.0%, and an AUC-ROC of 0.942 on the held-out test set, validated via 5-fold stratified cross-validation (CV AUC-ROC: 0.936). The model was chosen for its interpretability and efficient real-time processing of mixed clinical data. UMAP was explored as a dimensionality reduction step but was found to substantially reduce predictive performance (accuracy: 69.8%, AUC-ROC: 0.741) and was therefore retained only as a visualisation aid; all models were trained on the full raw feature set. Pre-existing comorbidities including diabetes, chronic kidney disease, heart failure, asthma and malignancy are associated with persistent post-COVID complications including fatigue, dyspnoea, tachycardia and chronic cough. The application translates these predictive outputs into an accessible interface enabling middle-aged users to assess their risk profile and seek timely medical consultation. It is important to note that the outcome variable in the source dataset reflects mortality (died vs. recovered) rather than long-term post-acute sequelae, and this distinction must be considered in any deployment context. This tool is intended as a screening aid only and is not a substitute for clinical diagnosis. Future work should incorporate multi-model refinement, biomarker integration, and formal clinical validation.

Recommended Citation

Amarathunga, Himasha; Dhammearatchi, Dhishan; and Pollwaththage, Nimesh (2026) "A Web-Based Machine Learning Application for Classifying Post-COVID Risk among Middle-Aged Adults," Journal of Innovation in Science, Engineering and Technology: Vol. 7: Iss. 2, Article 3.
DOI: https://doi.org/10.66543/3084-858X.1112

First Pages

26

Last Page

35