Comparative Analysis of Classification Models for Stroke Prediction Using Machine Learning Algorithms

Stroke has become a significant global health challenge among the taking the lead to death and long-term disability. Early monitoring of stroke risk is important in the field of preventive healthcare and effective in clinical decisionmaking. The study helps in upgradation of ML (machine learning) classification models for stroke prediction using a structured data. The structural data was pre-processed through categorical encoding, normalization of continuous variables, and class imbalance handling with Synthetic Minority Oversampling Technique with Edited Nearest Neighbours (SMOTEENN). A data was split into train-test phase, followed by baseline benchmarking with Lazy Classifier to identify promising structures. Three models 1. Logistic Regression, 2. Ridge Classifier, 3. Calibrated Logistic Regression were selected for in-depth analysis as a best model for further processing. Hyperparameter optimization was conducted using Grid Search, and the tuned models were further combined into Voting and Stacking ensembles to enhance predictive capability. Efficiency was evaluated using accuracy, precision, recall, specificity, sensitivity, F1-score, and ROC-AUC. The Phase-1 results demonstrate that Grid Search-based tuning and ensemble strategies improve classification outcomes, establishing a strong methodological foundation for reliable, interpretable, and data-driven stroke prediction systems.