Evaluation of performance on preterm birth classification using single wavelet-based features of EHG signals

In this study, wavelet-based features of electrohys-terogram (EHG) recordings obtained from pregnant women are applied for preterm birth classification. The wavelet-based feature of EHG signal, Δι, is determined from a difference between the logarithms of variances of detail coefficients of EHG signal corresponding to two consecutive levels, i.e., level I and level 1 + 1. The performance on preterm birth classification using single wavelet-based features is reevaluated using the receiver operating characteristic (ROC) analysis. The computational results show that best performance on preterm birth classification is achieved with the area under ROC curve of 0.7386 using the wavelet-based feature Δ
₆
of EHG signals recorded after the 37th week of gestation.