Boyle, Patrick MYeh, Shu-Yi2025-08-012025-08-012025-08-012025Yeh_washington_0250O_28246.pdfhttps://hdl.handle.net/1773/53404Thesis (Master's)--University of Washington, 2025This study presents a machine learning framework designed for early prediction of ventricular tachycardia (VT) using single-lead electrocardiogram (ECG) signals collected from portable monitoring devices. Unlike prior studies that relied heavily on hospital-based data or incorporated additional demographic information, this work enhances applicability for out-of-hospital patient care. Three modeling approaches were explored: a Long Short-Term Memory (LSTM) model, 2D Convolutional Neural Networks (CNN) trained on spectrograms, and a Support Vector Machine (SVM) using features extracted by a Variational Auto-Encoder (VAE). Among these, the VAE-SVM model demonstrated superior performance, achieving an F1 score of 0.66 and a recall of 0.77. Explainable AI techniques, latent space traversal, and correlation analysis were applied to interpret model behavior and identify physiologically meaningful features associated with VT onset. These findings highlight a valuable opportunity for developing wearable-based VT detection tools that can be integrated into daily health monitoring systems.application/pdfen-USnoneElectrocardiogramMachine LearningPortable Health MonitoringVariational AutoencoderVentricular TachycardiaBioengineeringBioengineeringThesis