Impact of Cross-Validation on Machine Learning Models for Early Detection of Intrauterine Fetal Demise

Abstract

Intrauterine fetal demise in women during pregnancy is a major contributing factor in prenatal mortality and is a major global issue in developing and underdeveloped countries. When an unborn fetus passes away in the womb during the 20th week of pregnancy or later, early detection of the fetus can help reduce the chances of intrauterine fetal demise. Machine learning models such as Decision Trees, Random Forest, SVM Classifier, KNN, Gaussian Naïve Bayes, Adaboost, Gradient Boosting, Voting Classifier, and Neural Networks are trained to determine whether the fetal health is Normal, Suspect, or Pathological. This work uses 22 features related to fetal heart rate obtained from the Cardiotocogram (CTG) clinical procedure for 2126 patients. Our paper focuses on applying various cross-validation techniques, namely, K-Fold, Hold-Out, Leave-One-Out, Leave-P-Out, Monte Carlo, Stratified K-fold, and Repeated K-fold, on the above ML algorithms to enhance them and determine the best performing algorithm. We conducted exploratory data analysis to obtain detailed inferences on the features. Gradient Boosting and Voting Classifier achieved 99% accuracy after applying cross-validation techniques. The dataset used has the dimension of 2126 × 22, and the label is multiclass classified as Normal, Suspect, and Pathological condition. Apart from incorporating cross-validation strategies on several machine learning algorithms, the research paper focuses on Blackbox evaluation, which is an Interpretable Machine Learning Technique used to understand the underlying working mechanism of each model and the means by which it picks features to train and predict values.

Predicting Car Rental Prices: A Comparative Analysis of Machine Learning Models

“…We divided the training and test sets in a ratio of 8:2. Cross-validation was applied to RF to prevent overfitting [16], whereas dropout and early stopping were used in the deep learning process. Early stopping used 10% of the training dataset to monitor the validation loss during the training process, and, if the validation loss was improved for 5 epochs, the training process was stopped.…”

“…To address this, they randomly sampled and trained decision trees, which were also applied in our experiments. Cross-validation was referenced to prevent overfitting in RF and various ML models [16]. Raju et al used RF for accurate and data-driven rent prediction in the rapidly changing real estate market [25].…”

Parkinson Hastalığının İlerlemesini Tahmin Etmek: Ses Girişlerinden Yararlanan İnvazif Olmayan Bir Yöntem

“…The final performance metric, whether it's accuracy, mean squared error, or any other relevant measure, is calculated as the average across these 'k' iterations (White & Power, 2023). This method provides a more reliable estimate of a model's performance as it assesses its consistency across different data subsets, helping to identify potential overfitting or underfitting issues (Kaliappan et al, 2023).…”

Artificial Intelligence in Fetal Health Diagnosis: A Systematic Literature Review

“…using CV with OneR, ZeroR, RIDOR, and JRIP classifiers resulted in 99.74% accuracy for PNN and 99.25% accuracy for HECFNN using Alkhasawneh balanced the analysed dataset with SMOTE using Jrip, Ridor, J48, NBStar, IBk and Kstar classifiers, achieving 99.05% accuracy with IBK using CV (57,84,88). Kaliappan et al achieved 99% accuracy with GB and VC using various CV techniques such as Hold-Out, Leave-One-Out, Leave-P-Out, Monte Carlo, Stratified K-fold (25).…”

“…Kaliappan et al aimed to improve and determine the best performing algorithm among ML models, including decision tree (DT), RF, support vector machines (SVM), K-nearest neighbor (KNN), Gaussian Naïve Bayes, adaboost, GB, voting classifier (VC) and neural networks (NN), by applying various cross-validation (CV) techniques such as K-fold, Hold-Out, Leave-One-Out, Leave-P-Out, Monte Carlo, Stratified K-fold and Repeated K-fold (25). They used 22 features related to fetal heart rate obtained from the clinical CTG in 2126 patients.…”