Stroke is a life-threatening condition causing the second-leading number of deaths worldwide. It is a challenging problem in the public health domain of the 21st century for healthcare professionals and researchers. So, proper monitoring of stroke can prevent and reduce its severity. Risk factor analysis is one of the promising approaches for identifying the presence of stroke disease. Numerous researches have focused on forecasting strokes for patients. The majority had a good accuracy ratio, around 90%, on the publicly available dataset. Combining several preprocessing tasks can considerably increase the quality of classifiers, an area of research need. Additionally, the researchers should pinpoint the major risk factors for stroke disease and use advanced classifiers to forecast the likelihood of stroke. This article presents an enhanced approach for identifying the potential risk factors and predicting the incidence of stroke on a publicly available clinical dataset. The method considers and resolves significant gaps in the previous studies. It incorporates ten classification models, including advanced boosting classifiers, to detect the presence of stroke. The performance of the classifiers is analyzed on all possible subsets of attribute/feature selections concerning five metrics to find the best-performing algorithms. The experimental results demonstrate that the proposed approach achieved the best accuracy on all feature classifications. Overall, this study's main achievement is obtaining a higher percentage (97% accuracy using boosting classifiers) of stroke prognosis than state-of-the-art approaches to stroke dataset. Hence, physicians can use gradient and ensemble boosting-tree-based models that are most suitable for predicting patients' strokes in the real world. Moreover, this investigation also reveals that age, heart disease, glucose level, hypertension, and marital status are the most significant risk factors. At the same time, the remaining attributes are also essential to obtaining the best performance.
Key words: Stroke prediction;
machine learning;
classification;
feature selection;
stroke risk factors;
healthcare.
|
