Quantum Bayesian Neural Network for Uncertainty-Aware ECG Anomaly Detection

This study develops a Quantum Bayesian Neural Network for electrocardiogram anomaly detection by integrating multi-scale feature engineering, variational quantum learning, Bayesian parameter modelling, and physical quantum hardware evaluation. Using the MIT-BIH Arrhythmia Database, the proposed pipeline applies leakage-safe preprocessing, record-adaptive multi-window analysis, dependency-aware feature construction, and robust feature selection before quantum encoding. Angle and Dense-Angle encoding are systematically evaluated across four- and six-qubit circuits, two and four variational layers, and configurations with and without data re-uploading. Bayesian inference is incorporated by representing quantum parameters probabilistically and estimating posterior predictive anomaly probabilities through repeated parameter sampling. Among the evaluated configurations, Dense-Angle encoding with six qubits, four layers, and data re-uploading provides the strongest overall probabilistic perfor- mance, achieving an area under the receiver operating characteristic curve of 0.91, an area under the precision-recall curve of 0.90, recall of 0.88, and a Brier score of 0.13. A compact four-qubit Dense-Angle model is additionally executed on the IBM Marrakesh quantum processor to investigate predictive consistency under physical quantum noise. These results demonstrate that quantum encoding density, circuit architecture, data re-uploading, and Bayesian inference jointly influence anomaly-detection performance and provide a practical foundation for uncertainty-aware quantum learning for electrocardiogram analysis.