Evaluating machine learning-based stroke prediction models in imbalanced data challenges

A stroke occurs when a blood clot or bleeding in the brain leads to potential permanent damage in areas such as mobility, cognition, vision, or speech. Strokes are considered medical emergencies and can result in long-term neurological impairments or, in severe cases, death. The majority of strokes can be categorized as either ischemic embolic or hemorrhagic. Ischemic embolic strokes occur when a blood clot forms in another part of the body, often the heart, and travels through the bloodstream to block narrower arteries in the brain. A hemorrhagic stroke refers to a type of brain stroke characterized by the leakage or rupture of a blood vessel in the brain. Stroke affects a significant portion of individuals aged 65 and above and is the second leading cause of death globally. It can result in heart damage, often referred to as a "heart attack." Strokes incur substantial medical expenses, can cause long-term disabilities, and in some cases, may lead to fatalities. Approximately one stroke is fatal every four minutes; 80% of strokes can be lessened if there is early detection of stroke or prediction of its occurrence. The results show that all the algorithms have a considerably high level of stroke prediction accuracy. The AdaBoost has the highest accuracy score of 93.7%, the random forest has the second-highest accuracy score of 91.8%, and the logistic regression has the third-highest accuracy score of 89.9%, while the decision tree has the lowest accuracy score of 89.5%.