The Ensembles of Machine Learning Methods for Survival Predicting after Kidney Transplantation

Applied Sciences

Nataliia Melnykova¹  Nataliya Shakhovska¹  Yaroslav Tolstyak²  Valentyna Chopyak²  Michal Gregus ml³  Igor Yakovlev⁴  Rostyslav Zhuk⁵ 

[1] Artificial Intelligence Department, Lviv Polytechnic National University, 79013 Lviv, Ukraine;Clinical Immunology and Allergology Department, Danylo Halytskyi Lviv National University, 79010 Lviv, Ukraine;Department of Information Systems, Faculty of Management, Comenius University, Odbojárov 10, 814 99 Bratislava, Slovakia;Hospital Nephrology and Dialysis Department, Lviv Regional Clinical Hospital, 79010 Lviv, Ukraine;Surgery and Transplantation Department, Danylo Halytskyi Lviv National University, 79010 Lviv, Ukraine;

关键词: organ transplantation;    Kapplan-Meier method;    machine learning;    ensemble;    early risk prediction;   

DOI  :  10.3390/app112110380

来源: DOAJ

【 摘 要 】

Machine learning is used to develop predictive models to diagnose different diseases, particularly kidney transplant survival prediction. The paper used the collected dataset of patients’ individual parameters to predict the critical risk factors associated with early graft rejection. Our study shows the high pairwise correlation between a massive subset of the parameters listed in the dataset. Hence the proper feature selection is needed to increase the quality of a prediction model. Several methods are used for feature selection, and results are summarized using hard voting. Modeling the onset of critical events for the elements of a particular set is made based on the Kapplan-Meier method. Four novel ensembles of machine learning models are built on selected features for the classification task. Proposed stacking allows obtaining an accuracy, sensitivity, and specifity of more than 0.9. Further research will include the development of a two-stage predictor.

【 授权许可】

Unknown