A Study for Predicting Cerebral Stroke Using Different Kind of Machine Learning Techniques

Abstract

An interruption or reduction in the supply of blood to the cerebrum produces a cerebral stroke. This supply shortage leads in a deficit of oxygen or vital nutrients, which causes brain cells to die. Stroke occurs mostly as a result of people's lifestyle choices in the advanced time-changing factors, for example, excessive glucose levels, heart disease, stoutness, and diabetes. Developing countries account for 85 percent of all stroke deaths worldwide. The early termination of a cerebral stroke is critical for effective counteraction and therapy. The best way to deal with this risk is to prevent it from happening in the first place by managing the relevant metabolic factors. Nonetheless, it is difficult for clinical workers to determine how much additional safety precautions are necessary for an expected patient based only on the examination of physiological indicators unless they are plainly abnormal. Examination reveals that behaviors extricated from various hazard limits transmit critical information for the prediction of stroke. The data was obtained from the Harvard Dataverse Repository and was properly prepared and tested using machine learning techniques such as RF, LR and KNN. We implemented the RF, LR, and KNN algorithms with hyperparameter tuning in this study and determined the best method among them. For performance evaluation, we use the AUC-ROC curve, the Precision-Recall curve, and the F1-score, and all reports reveal the best strategies. This strategy may be seen as a different option, with a low cost and a constant analytic technique that can get exact stroke prediction.