Comparative Analysis of Ensemble Learning Methods for Enhancing Fetal Health Prediction Using Cardiotocography

Abstract

Nearly half of the world’s stillbirths occur during labour and delivery. Early detection of any fetal distress can prompt the doctors to take appropriate measures. Cardiotocography (CTG) is one such technique that continuously records the fetal heart rate and uterine contractions during childbirth. Along with indicating signs of fetal hypoxia, CTG can also be interpreted to detect fetal abnormalities. Using the cardiotocography dataset from the UCI Machine Learning Repository, our paper displays a comparative analysis of different classifiers and ensemble learning methods such as max voting, weighted average, blending, bagging and boosting to enhance the fetal state prediction. Of all the ensemble methods used in our analysis, it was found that the Light Gradient Boosting Machine (LightGBM) gave the highest accuracy of 95.90%, which exceeded similar existing models. This increase in accuracy can prove to be potentially life saving, aid doctors in a more accurate detection of fetal abnormalities, reduce human error rates and increase infant mortality rates.

Fetal Hypoxia Detection Using Machine Learning: A Narrative Review

“…Nonetheless, the prevalent reliance on a single dataset in studies [25][26][27][28] highlights a significant research gap, underscoring the imperative of employing more diverse datasets to substantiate these models' validity. In the context of ensemble techniques, especially Gradient Boosting (GB) models, studies [41,[43][44][45] have yielded outstanding results. Yet, the repeated use of similar datasets, with study [44] being an exception as it utilized a unique private dataset, raises concerns about the diversity of datasets.…”

“…Research by Chidambaram and Joy [43] aimed to analyze different classifiers and ensemble techniques to enhance fetal health prediction. In the study, a dataset similar to [9] was used.…”

“…Based on an extensive literature review, previous studies have primarily concentrated on identifying and preventing fetal hypoxia. However, it is worth noting that the datasets presented in some of the papers, such as [11,18,19,21,23,31,35,36,[38][39][40]43], have limited data and imbalanced samples, which may have affected the study's bias. Additionally, in [29], the analysis was restricted to visual data because the dataset only contained images.…”

“…An additional significant deficiency observed in the research's findings was the inadequate management of data preprocessing procedures. For instance, the handling of noise present in the dataset is not addressed in the research paper [43]. Data augmentation is used in the study [46], which raises the possibility of increasing label noise in the dataset.…”