A Refined Artificial Neural Network Framework for Superior Fetal Health Monitoring

Abstract

Perinatal mortality is a dismal aspect of fetal health over the world. Fetal health is a critical component of a sustainable nation. Cardiotocography (CTG) employed in the fetal health monitor to assess mother and fetal well-being during childbirth. Cardiotocograms monitor the heart rate at its baseline value during labor, ignoring the risk of perinatal before delivery. The classifier predicts CTG data, allowing obstetricians to observe and anticipate the fetus status. This classifier identifies fetal normal, suspicious and pathologic states, with the heart rate of baseline value involved indicating a difficult fetal. A CTG dataset contains 2126 recording from the UCI repository used for classification. The CTG data sets are preprocessed using Normalization and scaling techniques, and then the preprocessed input data is utilized to Artificial Neural Network (ANN) model construction for classifying the fetal health, which is normal, suspicious, and pathologic of the high risk of pregnancy during 32 and 33 weeks of observation. Further, the 5-fold cross-validation is applied to examine the proposed model is quantified by precision, recall, F-score, ROC curve generated for three classes. From the evaluation, the Tuned ANN attained accuracy as 92% outperforms the un-tuned ANN accuracy of 90%.