Stroke Detection System Using Machine Learning

Abstract

Stroke is one of the leading causes of death and long-term disability worldwide, making accurate and timely risk prediction important for improving patient outcomes and supporting preventive healthcare. This paper presents a comprehensive machine learning-based stroke detection system that uses clinical and demographic patient information for probabilistic risk assessment. The system was developed using the Kaggle Stroke Prediction Dataset containing 5,110 patient records and 11 clinical features. To address class imbalance, the Synthetic Minority Over-sampling Technique (SMOTE) was applied to the training data, while missing BMI values were handled using median imputation, followed by feature encoding and scaling. Six supervised classification algorithms—Random Forest, Decision Tree, Support Vector Machine (SVM), Logistic Regression, Multi-layer Perceptron (MLP), and K-Nearest Neighbour (KNN)—were trained and evaluated. Feature selection was performed using correlation analysis followed by Recursive Feature Elimination with Cross-Validation (RFECV). On the held-out test set, the Random Forest classifier achieved the best overall performance, with an accuracy of 95.60%, precision of 70.00%, recall of 14.29%, F1-score of 23.73%, and ROC-AUC of 0.8465. The proposed framework further integrates SHAP-based patient-level explainability and a Flask REST API with healthcare interoperability support. The system provides a modular framework for stroke risk prediction and can be further extended toward IoT-based monitoring, federated learning, multimodal clinical data integration, and prospective clinical validation.