<p>Monitoring the fetus's health during pregnancy is crucial to preventing abnormalities that may harm the pregnant woman and the unborn child. In pregnancy, cardiotocography (CTG) is the most used technique for determining the heartbeat of the infant&#xa0;and the pattern of contractions in the uterus. Doctors rely on these data to assess the risks to both the mother and fetus to provide clinical suggestions. Nevertheless, traditional CTG data processing consumes a lot of time and may result in improper fetal health assessments. To address these issues, a Bidirectional Gated Recurrent Unit Neural Network with Dropout regulation based on the Water Wheel Plant Algorithm (W-BiGRU-D) evolved to efficiently classify the health status of the fetus using CTG data. The first step in the proposed model is to gather CTG data, which includes data regarding fetal movements, uterine contractions, and Fetal Heart Rate (FHR). The gathered dataset is preprocessed using modified GAN (Generative Adversarial Network) imputation to fill in missing value and maximum absolute scaling for standardization thus improving data quality. Next, the pre-processed data is subjected to the Relief F-MRFE (Modified Recursive Feature Elimination) feature selection model, which enhances the accuracy of the classifier model by selecting the most significant feature for the fetus health classification. The selected data is subsequently fed into a W-BiGRU-D classifier, which determines if the fetal health is normal, suspicious, or abnormal. To improve model performance in the BiGRU-D classifier, the Water Wheel Plant algorithm (WWPA) is used to optimize hyper-parameters including the learning rate and hidden layer. Performance evaluations of 98.59% accuracy, 96.58% precision, and 95.24% recall were obtained using the proposed approach. Based on the results, the proposed model performs superior to existing methods. Thus, the proposed optimized deep learning approach effectively categorize fetal health state with higher accuracy.</p>