Monitoring fetal well-being during pregnancy is an essential task which helps to evaluate the baby's growth and development. Proper and continuous monitoring addresses the needs of both mother and baby before, during, and after the birth. Fetal mortality, especially Perinatal mortality, is one of the major problems of fetal health monitoring. Perinatal mortality is the main cause of death in preemie. Hence, early diagnosis of fetal health is significant for any sustainable fetal health care environment. During pregnancy, Cardiotocography is used to monitor fetal health, judge maternal and fetal well-being. Cardiotocography (CTG) records the fetal baseline heartbeats per minute, fetal movements per second, uterine contraction per second, etc. Using these CTG records, obstetricians monitor the state of the fetus. Early diagnosis can be done by developing a predictive model using these records. Artificial Neural Networks (ANN) is a computing system used to build a model for classification or prediction problems. Compared to other classification models, ANN proves to have better results in healthcare and medical applications. Hence in this work, ANN is used to build the classification model using the CTG dataset consisting of 2126 records from the UCI repository. The proposed ANN model predicts the state of a fetal is normal, suspicious or pathologic. Early diagnosis using ANN decreases the high risk of pregnancy during 38 weeks of observation and doctors can accurately predict the complications in the pregnancy. Further, the performance of the ANN is improved by fine-tuning its hyper-parameters using Grid Search Cross-Validation Technique (GSCVT), and Randomized Search Cross-Validation (RSCVT). Fine-tuned ANN model with GSCVT, RSCVT achieved 94.11% classification accuracy, which is superior to Un-tuned ANN.

Cardiotocogram, Fetal health, Baseline value, Uterine Contraction, standardization techniques, Artificial Neural Network, Grid search cross-validation technique, Random search Cross-validation technique.