Answers are grounded in this paper and cite their sources.
The aim of the study was to evaluate the performance of SMOTE, GAN and VAE data synthesis methods in the task of predicting postoperative atrial fibrillation (PoAF) and in-hospital mortality (IHM) in coronary heart disease (CH) patients after coronary artery bypass grafting (CABG).Materials and methods. A single-center retrospective study was conducted, in which the medical history data of 999 patients with CHD undergoing elective CABG were analyzed. The end points of the study were PoAF and IHM. Development of predictive models was performed using machine learning methods: multivariate logistic regression (MLR), random forest (RF) and eXtreme Gradient Boosting (XGB). Nine data synthesis methods were used to generate new minority class samples: 5 SMOTE group methods, SOMO, GAN, WGAN and VAE methods.Results. Comparison of quality criteria for the predictive models of PoAF and IHM, developed on the basis of real and synthetic data, showed that for the MLR and RF models, the use of synthetic objects was not associated with an increase in prediction accuracy. When using the XGB method to solve IHM prediction problem, in which the majority class volume was dominant (15 to 1), only the ProWRAS method was associated with an increase in prediction quality. When class imbalance is not significant (4 to 1), which corresponds to the PoAF end point, the use of data synthesis methods does not improve prediction quality.Conclusion. The use of SMOTE, GAN and VAE methods does not guarantee an improvement in the accuracy of predictive models for PoAF and IHM in CHD patients after CABG
No takes yet. Share an insight, caveat, or question.
Shakhgeldyan et al. (2024) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: