Classification of Fetal Cardiac Arrhythmia Using Heart Rate Variability and Machine Learning

Abstract

This thesis presents a comprehensive study on fetal arrhythmia detection and classification using Heart Rate Variability (HRV) features and machine learning models. The dataset obtained from the NIFEEG Database on PhysioNet was employed for evaluation. Machine learning classifiers, including Support Vector Machine (SVM), k-Nearest Neighbors (KNN), Decision Tree, XGBoost, Naive Bayes, and Convolutional Neural Network (CNN), were analyzed for their efficacy in detecting fetal arrhythmias. The analysis of time domain, frequency domain, and non-linear HRV parameters provided valuable insights into fetal heart rate patterns. While SVM, KNN, Decision Tree, and XGBoost showed promising results, CNN emerged as the top-performing model, demonstrating high accuracy of 98.16%, sensitivity of 100%, specificity of 96.15%, and F1-score of 98.11% during validation and testing, indicating its potential for early detection of fetal arrhythmias. The findings support the significance of HRV features in fetal arrhythmia detection, paving the way for improved prenatal care and fetal health monitoring. Future directions emphasize the need to expand the dataset size to improve model generalizability. Additionally, exploring advanced deep learning architectures and integrating supplementary features, such as uterine contraction information, can further enhance model performance. In conclusion, this study showcases the potential of HRV-based machine learning models for fetal arrhythmia detection. The CNN model's outstanding performance offers valuable clinical implications for early detection and management of fetal arrhythmias.