Abstract:
A stroke can happen when the blood supply to a certain region in the brain’s cortex is abruptly severed. Without adequate
blood flow, brain cells will eventually die, and the extent of the damage will be inversely correlated with the area of the injured
brain. Early symptom identification is essential for stroke prediction and encouraging healthy habits by giving helpful information.
The solution to these issues is developing a precise and effective early-stage prediction model employing analytical support in clinical
decision-making with digitized patient information. Most research focuses on forecasting cardiovascular stroke, but the cerebral risk
for stroke has gotten far less attention. The current study aims to progress and assess several machine learning models to create a
framework for predicting the long-term potential of cerebral stroke. By conducting a thorough experimental assessment using two
different methodologies, namely SMOTE and SMOTE ENN, the suggested work tried to address the issue of imbalanced data from the
Kaggle dataset. On SMOTE-Balanced as well as SMOTE ENN-balanced datasets, several models such as K-nearest neighbour (KNN),
logistic regression(LR), support vector machine(SVM), decision tree(DT), random forests, XG boosting, stacking, and ANN are trained.
As demonstrated by the results, SMOTE ENN and the stacking classification approach achieved a remarkable 99.52% accuracy, with
a recall of 99.30%, a precision of 99.4%, and an F1 measure of 99%. We found that a relatively simple data balancing technique
combined with a supervised machine learning algorithm can be used to predict strokes with high accuracy and great practical potential.
