Ischemic Stroke Prediction Utilizing a Class Balancing Approach and Performance Comparison using Different Machine Learning Algorithms

Cerebrovascular disease often known as stroke, is the leading induce of death and permanent disability. Early detection of the undefined helps prevent its consequences. Stroke is widely known by its category, whether it happens due to the bursting of blood vessels or when a plague-like clot deposits in the arteries, which is known as ischemic or haemorrhagic stroke, therefore it's expected to predict the problem as soon as possible. In this study, the International Stroke Trail (IST) dataset is utilized and the size of the dataset is 19435 rows and 112 columns in total, but only essential features that are relevant to the study being considered. The collected data is subjected to data pre-processing so that the raw data is turned into feasible data, where data imputation, data labeling, and other pre-processing techniques are conducted, and it is also discovered that data suffers from class imbalance, which is inevitable in clinical datasets. The study aims to determine how using skewed information might lead to good accuracy that deceives the model's robustness in order to deal with this issue SMOTE (Synthetic Minority Oversampling Technique) is utilized. Early detection of ischemic stroke is recommended so in this study contemplate machine learnedness algorithms such as random forest, decision tree, xgboost, and binary logistic regression is utilized in that random forest with SMOTE obtained 76.4% accuracy and other evaluation metrics like precision, recall, f1 score are also measured. The performance of the testing and training data is done in IBM SPSS (Statistical Package for the Social Sciences)