Stroke Prediction System Using Machine Learning Methods

Abstract

Stroke, a deadly disease affecting the brain, has damaging outcomes which may result to death. Its burden has significantly increased in developing countries due to the lack of resources focusing on stroke healthcare and prevention. The need to minimize its effects surged the need to be cautious against the disease and use digital instruments to improve identifying stroke risk. This study implemented different machine learning techniques to predict the probable occurrence of stroke. After removing noise and outliers, data pre-processing was applied along with KNN Imputation to impute missing values. SMOTE was used to handle the imbalance present in the data and after conducting feature selection with the use of ExtraTreesClassifier, XGBoost generated the highest performance metrics among the 7 classifiers. The model was then integrated to the web application making it possible for users to predict whether or not they have the likelihood of having the disease.