Correctly estimating fetal birth weight (FBW) before delivery is a clinically important problem, as it is directly associated with perinatal mortality and long-term developmental outcomes. Low birth weight (LBW, <2500 g) and macrosomia (HBW, >4000 g) are associated with increased maternal and neonatal complications during pregnancy and a higher probability of chronic diseases in later stages of life. Standard sonographic regression formulas demonstrate systematic errors for extreme weight categories due to cross-institution and intergroup morphological heterogeneity in clinical datasets. Current machine learning (ML) methodologies generally depend on cardiotocography (CTG) signals or specialized ultrasound biometric features (biparietal diameter, femur length, abdominal circumference), which are often inaccessible in primary care antenatal environments. This paper presents VeraWeight, a heterogeneous stacking ensemble that integrates predictions from Random Forest (RF), Gradient Boosting (GB), and Logistic Regression (LR), trained on a specific set of vector-missing maternal clinical attributes collected at each antenatal visit. The preprocessing pipeline involves six stages: validation steps, handling missing values using the median, removing outliers based on IQR, one-hot encoding, Min-Max normalization, and train-test split. This enhances data accuracy. Experimental results confirm that VeraWeight achieves an overall accuracy of 95.45%, a macro-average F1 score of 0.95, a Mean Absolute Error of 193.7, and an RMSE of 264.1. It outperforms each individual base learner with statistical significance (McNemar’s test, p < 0.05).

We demonstrated VOTE-WEIGHT, an algorithm that combines models that can vote in their own way—Random Forest [16] and Gradient Boosting [17]. The framework also includes Logistic Regression to classify outcomes into three groups. Doctors can predict a baby’s weight using 12 standard checkup parameters, following a systematic sequence of preprocessing steps before analysis (Algorithm 1). Through complete inspection of individual modules and subsequent refinement, we identified areas needing attention (see Table IV). The normalized dataset setup (Fig. 2) and the results align with current best methods. As demonstrated in Table V, the model achieved: Accuracy: 95.6% Macro-average F1 score: 0.95 The average error score of 193.7 beats all previous baseline models. McNemar’s test (p < 0.05) confirms statistical significance. Across all case studies, matching shows consistent patterns. Analysis of CTG ensemble configurations suggests that voting-based approaches, combined with robust preprocessing, perform effectively in predicting maternal risks. Recent updates indicate improved handling of: Missing values Body weight patterns Noise removal The model improves outcomes by focusing on risk screening for low birth weight anomalies in broader populations. At the primary care level, VOTE-WEIGHT performs comparably to WHO standards. Predicting infant health at birth will soon become a reality.

[1] M. Feng, L. Wan, Z. Li, L. Chen, and X. Qi, “Fetal Weight Estimation Via Unsupervised Deep Machine Learning,” IEEE Access, 2018. [2] M. S. Alam, A. S. M. K. Ahmed, and G. Hogue, “A Comprehensive Approach for Fetal Health Prediction Using Machine Learning and Ensemble Models,” 2021. [3] M. F. Kashef, L. R. Adam, and V. Chandra, “Fetal Health Classification and Vectors for Low Birth Weight Condition Monitoring,” 2020. [4] S. A. Ali, Y. B. Chen, and T. V. Hemantha, “Machine Vision Techniques for Fetal Disease Detection,” 2019. [5] N. Rahman, R. Hossain, M. Jahan, and R. Yousuf, “Comparison of Machine Learning Algorithms for Fetal Health Classification,” 2020. [6] H. Vaidya and A. Singh, “A Machine Learning-Based Prediction Model for Fetal Health Assessment,” 2021. [7] A. Mirmozaffari, E. Barmaki, and V. Tayan, “Fetal Health State Prediction Using Machine Learning Techniques,” 2018. [8] Y. Zhang and X. Zhai, “Fetal State Assessment Based on Multimodal Data,” IEEE, 2017. [9] Y. Cheng et al., “Fetal Magnetocardiography Classification Using Deep Learning,” IEEE, 2020. [10] A. Keras, E. J. Hamilton, H. Parry, and H. E. Kramer, “Time-Series of Neural and Dynamic Feature-Based Systems,” 2019. [11] G. Casillas, D. Salinas, and P. Gomez, “Predicting Risk in Neonatal Systems,” 2016. [12] J. Spita et al., “Early Detection for Fetal Heart Rate Classification,” 2012. [13] J. C. Haldane, R. S. Shannon, R. L. Tola, and C. K. Puri, “Evaluation of Birth Weight Using Ultrasound and Clinical Data,” 2014. [14] M. J. Hopwood, A. Elisabeth, R. Balakrishnan, and J. C. Williams, “Assessment of Neonatal Weight Using Clinical Parameters,” 2012. [15] J. Gonzalez, “A Systematic Review of Fetal Weight Prediction,” 2010. [16] L. Breiman, “Random Forests,” Machine Learning, 2001. [17] J. H. Friedman, “Greedy Function Approximation: A Gradient Boosting Machine,” 2001. [18] H. V. Chandra, E. B. Joseph, J. C. Shah, and W. P. Eggleston, “SMOTE: Synthetic Minority Over-sampling Technique,” 2002. [19] A. Friedman and R. Tibshirani, “Regularization Methods in Statistical Learning,” 2005. [20] J. L. Ramsay and C. J. Williams, “Measuring Disease Risk Using Ensemble Models,” 2007. [21] Q. McNemar, “Tests on the Sampling Error of Differences Between Correlated Proportions,” 1947.

© 2025 International Journal of Engineering and Techniques (IJET).

Digital Object Identifier (DOI)

IJET assigns a unique DOI to every accepted article via Zenodo for persistent linking and citation. Your article’s DOI: https://doi.org/{{doi}}

Open DOI About Zenodo DOI

Close