Electrohysterography based Preterm Birth Classification in the CEEMDAN Domain

Abstract

Preterm births are on the rise on a global scale, with premature babies suffering from severe health complications. Manual classification is time consuming and often erroneous, and the economic costs of preterm births are huge. Electro-hysterography, the uterine muscle signals, of pregnant women has proven to be useful for automated classification of preterm births. But most of the automated methods reported in the literature use oversampling techniques to overcome the class imbalance problem of the datasets. The few techniques which use balanced dataset without the use of oversampling often give moderate accuracy. In this work, we present the use of CEEMDAN by taking equal number of term and preterm records for carrying out the classification process. Several classifiers are implemented and comparison of the results shows that Extremely Randomized Trees classifier produces the best results. There has been significant improvement in classification accuracy compared to other studies conducted using similar number of recordings.