Leveraging AdaBoost and CatBoost to Classify the Likelihood of Brain Stroke

Abstract

Brain Stroke occurs when the blood flow to a portion of the brain is reduced or stopped, denying the brain's tissue nourishment and oxygen, which results in brain cell death. Many lives can be saved by early diagnosis, but the bulk of clinical datasets, including the stroke dataset, are unbalanced, which means that the majority of predictive algorithms are biased. By balancing the dataset, resampling methods improve machine learning algorithms' capacity for prediction. This study compares various algorithms on a stroke dataset to determine the likelihood of experiencing a stroke. In order to predict stroke, the authors of this work used two machine learning classifiers, AdaBoost and CatBoost, in conjunction with a well-known resampling technique called Synthetic Minority Oversampling Technique (SMOTE). A publicly available dataset was employed for the study. CatBoost outperformed AdaBoost and achieved an accuracy of 96 % when combined with SMOTE. The accuracy achieved using CatBoost was better than that of most previously developed models and is on par with other advanced models.