Implementation of Various Machine Learning Algorithms for Prediction Analysis of Fetal Health Incorporating Feature Retraction

This study investigates the performance of various machine learning algorithms in the prediction of fetal health, emphasizing the impact of feature retraction on model accuracy and efficiency. Five classifiers Naive Bayes, K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Random Forest, and Logistic Regression were evaluated both with and without feature retraction. Performance metrics including accuracy, precision, recall, and F1-score were analyzed alongside visual results from confusion matrices. Without feature retraction, Random Forest achieved an accuracy of 96%, while Logistic Regression and KNN recorded 88% and 90%, respectively. However, when feature retraction was applied, both Random Forest and Logistic Regression achieved 100% accuracy, indicating an increase of 4.17% and 13.63%, respectively. KNN’s performance improved slightly from 90% to 90% across all metrics, maintaining stable results. In contrast, SVM’s accuracy dropped from 91% to 80%, highlighting the model’s dependency on complete feature sets. Naive Bayes showed marginal improvement in precision and F1-score, with F1 rising from 0.82 to 0.83. The confusion matrices further validated these findings. Before feature retraction, Logistic Regression misclassified 24 samples; post-retraction, it achieved flawless classification. Similarly, Random Forest reduced its misclassifications from 9 to 0, reflecting a 100% improvement in class prediction accuracy. These results demonstrate that applying feature retraction significantly improves model precision and generalization, particularly for tree-based and linear models. This study underscores the critical role of feature engineering in enhancing fetal health prediction, providing a foundation for deploying optimized ML models in clinical decision support systems.