An Imbalanced-Data Processing Algorithm for the Prediction of Heart Attack in Stroke Patients

Early predicting heart attack out of stroke patients in a view data analysis is an approach to reduce high mortality rate. Stroke-patient Intensive Care Unit are imbalanced due that with the minority patients. How predict stroke-patient becomes challenge. For processing data, this paper designs algorithm by leveraging random undersampling, clustering and oversampling techniques, which called undersampling-clustering-oversampling (shortly, UCO algorithm). The generates nearly balanced utilized train machine-learning models for attack. Over database Medical Information Mart III, extensive experiments conducted evaluate algorithm. A setting undersampling number 120 UCO, denoted UCO(120), shows good performance helping classifiers extract features. Five separately deployed based on outputs UCO(120). Our results show forest classifier achieves best accuracy 70.29%, precision 70.05%. It could be well-predicted using UCO(120) whether patient will have or not.