Machine Learning Approaches to Stroke Prediction Based on the Framingham Cardiovascular Study Dataset

Abstract

Stroke is the second biggest cause of death and long-term paralysis globally. It continues to be a significant health burden for both the elderly and national healthcare systems. Hypertension, heart illness, atrial fibrillation, diabetes, and other aspects of one's lifestyle are all potentially modifiable risk factors for a stroke. Then, putting machine learning concepts into practice over an existing health study dataset to effectively and accurately predict the occurrence of stroke will help with early intervention and treatment. In this study, we propose various machine learning methods for stroke prediction and compare them to available methods or approaches from other similar studies. Furthermore, we present a Naive Bayes probabilistic method that combines the concept of data imputation, class imbalance, and feature selection for stroke prediction, which achieves a greater area under the ROC curve than the Multilayered Perceptron neural network and the SVM proposed as the baseline methods for stroke prediction. In addition, neurologists can use our work to identify potential risk factors for stroke without any clinical trials methods. Finally, our methods can be applied to the clinical prognosis of other diseases, where data are often lacking and risk factors are poorly understood.