Background: Stroke is a leading cause of mortality and long-term disability worldwide. Early prediction of stroke can significantly enhance preventive measures and medical interventions. Extreme Gradient Boosting (XGBoost) has emerged as a powerful machine learning tool due to its robustness and efficiency. This study aims to develop and validate a stroke prediction model using XGBoost, incorporating various clinical, demographic, and lifestyle factors. Methods: A case-control study was conducted at a tertiary care hospital in Madhya Pradesh, India, involving 1360 participants aged 18 years and above. Patients diagnosed with stroke based on standard clinical criteria formed the case group, while the control group included individuals suspected of having a stroke but whose final diagnosis were negative. Data on various risk factors were collected and analysed using R 4.3.1 software. The dataset was divided into a training set (70%) and a testing set (30%). XGBoost, Random Forest, Logistic Regression, and Support Vector Machine (SVM) models were trained and evaluated using metrics such as accuracy, precision, sensitivity, specificity, F1 score, and AUC-ROC. Resampling techniques were applied to address dataset imbalances. Results: XGBoost demonstrated superior performance compared to other models. Without resampling, XGBoost achieved an accuracy of 91.50%, precision of 90.60%, and an AUC of 90.00%. With resampling, the performance improved, with an accuracy of 94.00%, precision of 94.00%, and an AUC of 95.00%. Random Forest also performed well, achieving an accuracy of 92.09% and an AUC of 91.93% without resampling. Other models showed lower performance metrics. Conclusion: XGBoost is a highly effective tool for stroke prediction, outperforming other machine learning models. Integrating clinical, demographic, and lifestyle factors into the XGBoost model enhances its predictive accuracy, making it valuable for early stroke detection and intervention. Future research should focus on refining these models and validating them on external datasets.