Secure Web-Based Early Stroke Detection: A Machine Learning Approach with Explainable Insights

Benson Mansingh, Neeraj Kurapati, Vasim Akram Shaik, Sindhu Madhav Bollu, Ruchitha Sure

2025

Abstract

Stroke of both ischemic and hemorrhagic origins continues to be a major public health burden worldwide, and there is an increasingly urgent need for novel strategies to identify stroke risk and prevent occurrence. In this work, we propose a novel secure web-based predictive system incorporating novel application of state-of-the-art machine learning coupled with clinical interpretability aspects, which would enable the healthcare professional. The model uses XG Boost, Random Forest, and k Nearest Neighbors (KNN) so that it can analyse vital health-related data, including age, hypertension, and glucose levels to deliver personalized stroke risk assessments. To counter data imbalances in stroke datasets, the Synthetic Minority Oversampling Technique (SMOTE) was applied, creating equal representation of stroke & non-stroke cases during training. More systematic hyperparameter optimization demonstrated that XG boost was indeed the best model, correctly classifying 93.2% of the 37,443 visible galaxy wings we assessed, outperforming the other classifiers. In addition to predictive performance, the system also puts interpretability first to encourage trust from the clinic. The model uses SHAP (Shapley Additive Explanations) and LIME (Local Interpretable Mode lagnostic Explanations) to explain its reasoning, highlights important risk drivers (e.g., age, high blood pressure) as well as individual explanations for patients. Predictive accuracy, born of complex algorithms, and approachability, granting easy access to patient-specific, actionable insights, come together in a virtuous cycle of profound clinical utility and wide adoption, enabling the timely, precise, and personal delivery of interventions. Besides its predictive potential, the webapp is implemented with strong security features. HTTPS encryption has been the best way to protect user data and secure communication. Additionally, a TaskLimiter module is also added to protect the system from Distributed Denial-of-Service (DDoS) attack in case of high traffic scenarios. This novel research goes beyond conventional statistical model by incorporating advanced machine learning, interpretable AI methods and strong security techniques to enable stroke prediction. The system’s high accuracy, transparency, and security underscore its potential for real-world deployment, contributing to proactive, data-driven healthcare strategies aimed at reducing strokerelated morbidity and mortality globally.

Paper Citation

in Harvard Style

Mansingh B., Kurapati N., Shaik V., Bollu S. and Sure R. (2025). Secure Web-Based Early Stroke Detection: A Machine Learning Approach with Explainable Insights. In Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies - ICRDICCT`25; ISBN 978-989-758-777-1, SciTePress, pages 429-437. DOI: 10.5220/0013914300004919

in Bibtex Style

@conference{icrdicct`2525,
author={Benson Mansingh and Neeraj Kurapati and Vasim Shaik and Sindhu Bollu and Ruchitha Sure},
title={Secure Web-Based Early Stroke Detection: A Machine Learning Approach with Explainable Insights},
booktitle={Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies - ICRDICCT`25},
year={2025},
pages={429-437},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0013914300004919},
isbn={978-989-758-777-1},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies - ICRDICCT`25
TI - Secure Web-Based Early Stroke Detection: A Machine Learning Approach with Explainable Insights
SN - 978-989-758-777-1
AU - Mansingh B.
AU - Kurapati N.
AU - Shaik V.
AU - Bollu S.
AU - Sure R.
PY - 2025
SP - 429
EP - 437
DO - 10.5220/0013914300004919
PB - SciTePress