International Journal of Scientific Research and Engineering Development

Stroke is a leading cause of death and long-term disability worldwide, responsible for approximately 5.5 million deaths annually. Early identification of high-risk individuals enables timely intervention and significantly reduces morbidity. This paper presents a comprehensive machine learning framework for binary stroke risk classification using electronic health record data. We evaluate six algorithms—Logistic Regression, Decision Tree, Random Forest, Gradient Boosting, AdaBoost, and KNearest Neighbours (KNN)—on the publicly available Stroke Prediction Dataset (n = 5,110). To address severe class imbalance (4.87% positive cases), a custom Synthetic Minority Over-sampling Technique (SMOTE) is implemented without external libraries. Three engineered features—age-glucose interaction, BMI-age ratio, and a composite risk score—are introduced to enrich the feature space. Hyperparameter optimisation is performed via 3-fold GridSearchCV. Models are assessed using ROC-AUC, Precision-Recall AUC, F1-score, sensitivity, and 5-fold cross-validated AUC. Logistic Regression achieves the highest test ROC-AUC of 0.834 with 80% recall, while Random Forest yields the highest cross-validated AUC (0.993±0.001). The complete pipeline—including a production- ready Streamlit web application and all reproducibility artefacts—is released publicly. This work contributes a reproducible, clinically interpretable baseline for stroke prediction research.

Shifa Bilal Tamboli, Simeen Phiroj Mulani, Arman Tajuddin Shiakh,"Stroke Risk Prediction Using an Ensemble Machine Learning Framework with SMOTE-Based Class Balancing" International Journal of Scientific Research and Engineering Development, V9(2): Page(2234-2239) Mar-Apr 2026. ISSN: 2581-7175. www.ijsred.com. Published by Scientific and Academic Research Publishing.