Realistic preterm prediction based on optimized synthetic sampling of EHG signal

Preterm labor is the leading cause of neonatal morbidity and mortality in newborns has attracted significant research attention from many scientific areas. The relationship between uterine contraction underlying electrical activities makes electrohysterogram (EHG) a promising direction for detecting predicting preterm births. However, due to scarcity EHG signals, especially those births, synthetic algorithms have been used generate artificial samples birth type order eliminate bias prediction towards normal delivery, at expense reducing feature effectiveness automatic detection based on machine learning. To address this problem, we quantify effect (balance coefficient) features form general performance metric by using several scores with relevant weights that describe their contributions class segregation. In combination activation/inactivation functions characterize abundance training accuracy obtained an optimal sample balance coefficient compromises removing toward majority group (i.e., delivery side importance features). A more realistic predictive was achieved through series numerical tests publicly available TPEHG database, therefore demonstrating proposed method.