DefinePK

Abstract

Premature delivery (PMD) refers to the occurrence of childbirth before 37 weeks of gestation. It is a significant worldwide health concern that can have adverse consequences on both infants and mothers. Models, including PMD-LSTM, PMD-GRU, PMD-Vanilla RNN, and PMD-ANN, are built and trained using the pre-processed PMD data. To forecast whether delivery would be preterm or non-preterm, a sigmoid activation function is employed in the output layer, together with binary cross-entropy loss, during the training process for binary classification. The performance of the model is evaluated using measures like loss, recall, accuracy, precision, and F1-score. The proposed models exhibit robust prediction abilities, with the LSTM achieving an accuracy of 0.6666, the GRU achieving an accuracy of 0.0166, the Vanilla RNN achieving a perfect accuracy of 1.0000, and the ANN achieving an accuracy of 0.9166. The metrics of precision, loss, recall, accuracy, and F1-score provide more understanding of the models' capacity to distinguish between preterm and non-preterm cases. This study contributes to the current research on employing neural network algorithms to detect preterm births at an early stage. This has the potential to result in timely interventions and improved outcomes for both moms and newborns. By comparing many architectures of Recurrent Neural Networks (RNNs), we can identify their respective strengths and weaknesses in addressing this critical healthcare problem.