Realisation of fetal health diagnosis with machine learning approaches

Several tests are used to assess the health of the foetus during pregnancy. The most reliable and widely used of these tests is the non-stress test, which measures the fetal heart rate and uterine contractions to assess fetal health. However, with traditional methods, detailed interpretation of this test takes time, there is variation between specialists, and interpretations can be misleading, especially by inexperienced specialists. These results lead to fetal losses and have a negative impact on perinatal mortality rates, which affect the level of development of countries. The aim of this study was to investigate an ensemble learning based approach to minimise diagnostic errors by assessing the fetal condition. In this direction, several experiments with different machine learning algorithms such as Linear Regression, Random Forest, Gradient Boosting and XGBoost were performed on cardiotocography data consisting of 22 features and 2126 samples. These experiments were performed on the dataset using principal component analysis and tenfold cross-validation methods. The performance of the algorithms used in the study was measured by accuracy, sensitivity, precision, F1 score and Gmean metrics. In the experiments, it was observed that the performance of the algorithms increased with the application of cross-validation and class weighting. In addition, the weighted majority voting method used in ensemble learning approaches has been found to increase the accuracy of the predictions obtained by combining the predictions of different machine learning models by weighting them. In this way, more reliable and consistent results are obtained than the predictions made by the algorithms alone. In the experimental results, it is observed that the accuracy rate of %99.99 obtained with the proposed XGBoost model is higher than previous literature studies. The weighted majority voting method will play a crucial role in reducing fetal mortality by improving diagnostic accuracy, especially in complex datasets and high-risk medical applications. The proposed model will help to reduce diagnostic errors and provide a more accurate assessment by automating fetal health analysis.