Integrative Multimodal Data-Driven Machine Learning Approach for Early Prediction of Stroke Risk and Severity Towards Personalized Prevention and Reduction of Stroke-Related Mortality

Ushasree R ^1,2*, Dr. Garima Sinha ³, Dr. Deepak Kumar Sinha ⁴

¹Research Scholar, SCSE, JAIN (Deemed-to-be University), Bangalore, India.
²Assistant Professor, Dept of MCA, Dayananda Sagar Academy of Technology and Management, Bangalore, India.
³Professor, SCSE, JAIN (Deemed-to-be University), Bangalore, India.
⁴Professor, SCSE, JAIN (Deemed-to-be University), Bangalore, India.

Mail Id: ^1,2*ushasreephdcse@gmail.com, ³mailatgarima@yahoo.co.in, ⁴dipu_sinha@yahoo.co.in

---

ABSTRACT

Stroke is one of the world's major causes of death and permanent disability, demanding early detection systems that go beyond conventional clinical risk scores. Traditional models often fail to capture the multifactorial nature of stroke, especially when relying on a single modality of data. This study presents an integrative Multimodal Machine Learning (ML) framework that fuses Electronic Health Records (EHR), neuroimaging, laboratory biomarkers, lifestyle indicators, and demographic factors to forecast the severity and risk of a stroke. The proposed framework applies ensemble learning, convolutional and recurrent neural networks, and attention-based fusion to synthesize heterogeneous datasets. Experiments conducted on benchmark datasets such as MIMIC-III, UK Biobank, and local hospital records achieved an AUC of 0.92, outperforming unimodal models by a significant margin. The model also demonstrated 86% accuracy in stratifying stroke severity (mild, moderate, severe), correlating strongly with clinical outcomes such as hospital stay and mortality. Feature interpretability via SHAP highlighted key predictors including age, blood pressure, lesion volume, CRP levels, and physical activity. The results show that multimodal, explainable ML models have the potential to advance personalized prevention strategies, enable real-time risk scoring, and ultimately reduce stroke-related mortality.

Keywords: Stroke, Electronic Health Records (EHR), neuroimaging, Machine Learning (ML), Ensemble Learning, Convolutional and Recurrent Neural Networks, and Attention-Based Fusion.

How to cite this article: Ushasree R, Sinha G, Sinha DK. Integrative multimodal data-driven machine learning approach for early prediction of stroke risk and severity towards personalized prevention and reduction of stroke-related mortality. Int J Drug Deliv Technol. 2026;16(7s): 131-143; DOI: 10.25258/ijddt.16.7s.17

Source of support: Nil.

Conflict of interest: None