Abstract:- Stroke has no proven cure and is a leading cause of death and long-term disability worldwide. Deep learning-based methods may outperform current models in predicting stroke risk, but require large amounts of properly labeled data. Due to strict privacy policies in health systems, stroke data is typically distributed in small chunks across multiple institutions. The World Health Organization (WHO) claims that stroke is the leading cause of death and disability worldwide. Early detection of various stroke warning indicators can reduce the severity of stroke. Several machine learning (ML) and deep learning (DL) models have been implemented to predict stroke probability. The study uses several physiological markers, machine learning, and deep learning techniques. Hybrid deep transfer learning, super vector machine (SVM), decision tree (DT) classification, random forest (RF) classification, CNN+LSTM, voting classifiers, and more. In order to train the above model and make accurate predictions. The algorithm with the highest accuracy rate for this task was CNN+LSTM. An open-access stroke prediction dataset was used to develop the method. The accuracy percentages of the models used in this study were significantly higher than previous studies, indicating a high degree of confidence in these models. Its robustness has been demonstrated by comparing multiple models, and the scheme can be derived from research analysis.