Prediction and Analysis of Stroke Risk Based on Ensemble Learning

With the development of science and technology, the application of data mining in medical field is becoming more and more popular. Machine learning methods also plays an important role in disease prediction. Stroke is characterized by high incidence rate, high disability rate, high mortality rate and high recurrence rate, and it is also likely to cause other kinds of complications. In this paper, each feature in the stroke dataset was analyzed in order to find out the factors affecting stroke and conducts classification and prediction research on whether there is a disease risk. Specifically, the PCA (Principal Component Analysis) algorithm is used to extract the main feature components of data, the SMOTE (Synthetic Minority Oversampling Technique) algorithm is used to adjust imbalanced feature categories. Traditional machine learning classification algorithms, such as decision tree, SVM(support vector machines), and various ensemble learning algorithms are used for the prediction of stroke risk, so as to study the relationship between stroke disease and each feature, and the classification prediction model, so that we can prevent strokes in time and reduce the risk of stroke. Among all the models, Bagging (Bootstrap aggregating) has the best performance with an ROC value of 0.97.