Machine learning-based fetal health prediction and development of smart web application

Figures

• 

  Figure 1. Barriers to prenatal care in low-resource settings identified from Heaman et al. (2015), Gangolli et al. (2005), and UNICEF Indonesia (2017).

• 

  Figure 2. Traditional vs. ML-based fetal health assessment: Comparison synthesized from Bertini et al. (2022), Ponsiglione et al. (2021), Mechael (2009), and Hawrami et al. (2026). ML advantages are most pronounced in resource-limited settings.

• 

  Figure 3. Key milestones in ML-based fetal health research.

• 

  Figure 4. Workflow of the proposed methodology process.

• 

  Figure 5. Training and testing data set distributions for machine learning and neural networks based algorithms.

• 

  Figure 6. Distribution of each features and outliers for three fetal classes (Class 1: Normal, Class 2: Suspect, Class 3: Pathological) are visualized by Boxplots.

• 

  Figure 7. Correlation heatmap illustrating pairwise relationships among all 22 CTG attributes.

• 

  Figure 8. Original distributions of all features across the whole dataset without any resampling and normalization.

• 

  Figure 9. Normalized distribution of all features after applying StandardScaler feature scaling technique.

• 

  Figure 10. Training and testing dataset distributions after applying SMOTEENN.

• 

  Figure 11. Bar plot showing the results of the Kruskal–Wallis H-test performed on each feature across the three fetal health classes: normal (Class 1), suspect (Class 2), and pathological (Class 3).

• 

  Figure 12. Multi-layer perceptron architecture.