Paroxysmal atrial fibrillation (PAF) screening by ensemble learning

Abstract

Ensemble learning is a method created by using different combinations of experts. Using different combinations has a great potential to get better results in pattern classification problems. In this study, ensemble learning was used for the aim of PAF screening, i.e. finding whether a person is PAF patient or not from his/her ectopic-free ECG records. Both hierarchical and parallel structures of ensemble learning were tried To train experts, k–fold cross validation and bootstrap sampling methods were used and their performances were compared. Four different types of classifiers were used as experts. Dataset used consists of electrocardiogram (ECG) records from both PAF patients and non-PAF subjects. The results obtained are presented in tables.